Terminal-Bench 4.0
Measures whether an agent can complete complex, multi-step professional tasks in a CLI environment (v4.0).
- Category
- Agentic coding
- Unit
- Percent (%)
- Higher is better
- Comparison groups
- 3total
- Currently available model rankings · Ranked results: 27
- All models (records): Groups: 11 · Entries: 58
- Evidence entries: 10
Rankings
Values with different source types (official, third-party, paper) or measurement conditions cannot be compared, so rankings are separated by comparison group (the same source, or sources verified to use the same conditions).
Rankings include only models currently available in Claude Code / Codex, excluding legacy models.
- View on rankings page (by metric)
- Cost balance (frontier and effort changes)
- Overall ranking (performance first)
official results (Accuracy vs Cost figure announced by Anthropic (Opus 5.5 and Sonnet 5.5 pages))
This group combines the same Accuracy vs Cost figure (Terminal-Bench 4.0 and FrontierCode v1.1 main set) on the Claude Opus 5.5 and Claude Sonnet 5.5 announcement page for each model. The scores and costs of all points appearing on both pages match. When the same point appears on both pages, only 1 row is included in the comparison group (the row from the other page is retained as evidence without a key), and the note from the other source is written in the conditions for that 1 row.
| Rank | Model | effort | Score | Source cost (metric) | measured/reported by | source |
|---|---|---|---|---|---|---|
| 1 | Claude Sonnet 5.5 | Max | 70.6% | $12.54Cost per attempt | Anthropic | Introducing Claude Sonnet 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5.5 · Max: 70.6% $12.54" |
| 2 | Claude Opus 5.5 | Xhigh | 66.4% | $7.35Cost per attempt | Anthropic | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "Opus 5.5 · Xhigh: 66.4% $7.35" |
| 3 | Claude Opus 5.5 | Max | 64.8% | $11.24Cost per attempt | Anthropic | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "Opus 5.5 · Max: 64.8% $11.24" |
| 4 | Claude Opus 5.5 | High | 64.2% | $3.88Cost per attempt | Anthropic | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "Opus 5.5 · High: 64.2% $3.88" |
| 5 | Claude Sonnet 5.5 | Xhigh | 61.5% | $5.30Cost per attempt | Anthropic | Introducing Claude Sonnet 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5.5 · Xhigh: 61.5% $5.30" |
| 6 | GPT-6 Astra | High | 57.9% | $7.21Cost per attempt | OpenAI | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor. measured by: the figure makes no distinction / the table footnote says GPT series figures are reported values from OpenAI; the figure's effort label "High" was mapped to high in the OpenAI API reasoning effort; original table footnote: "GPT-6 Astra and GPT-5.6 Sol figures are as reported by OpenAI"Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "GPT-6 Astra · High: 57.9% $7.21" |
| 7 | Claude Opus 5.5 | Med | 57.6% | $2.94Cost per attempt | Anthropic | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "Opus 5.5 · Med: 57.6% $2.94" |
| 7 | GPT-6 Astra | Xhigh | 57.6% | $7.48Cost per attempt | OpenAI | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor. measured by: the figure makes no distinction / the table footnote says GPT series figures are reported values from OpenAI; the figure's effort label "Xhigh" was mapped to xhigh in the OpenAI API reasoning effort; original table footnote: "GPT-6 Astra and GPT-5.6 Sol figures are as reported by OpenAI"Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "GPT-6 Astra · Xhigh: 57.6% $7.48" |
| 9 | GPT-6 Astra | Max | 56.7% | $10.35Cost per attempt | OpenAI | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor. measured by: the figure makes no distinction / the table footnote says GPT series figures are reported values from OpenAI; the figure's effort label "Max" was mapped to max in the OpenAI API reasoning effort; original table footnote: "GPT-6 Astra and GPT-5.6 Sol figures are as reported by OpenAI"Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "GPT-6 Astra · Max: 56.7% $10.35" |
| 10 | GPT-6 Astra | Med | 53.9% | $6.15Cost per attempt | OpenAI | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor. measured by: the figure makes no distinction / the table footnote says GPT series figures are reported values from OpenAI; the figure's effort label "Med" was mapped to medium in the OpenAI API reasoning effort; original table footnote: "GPT-6 Astra and GPT-5.6 Sol figures are as reported by OpenAI"Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "GPT-6 Astra · Med: 53.9% $6.15" |
| 11 | GPT-6 Astra | Low | 49.7% | $4.95Cost per attempt | OpenAI | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor. measured by: the figure makes no distinction / the table footnote says GPT series figures are reported values from OpenAI; the figure's effort label "Low" was mapped to low in the OpenAI API reasoning effort; original table footnote: "GPT-6 Astra and GPT-5.6 Sol figures are as reported by OpenAI"Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "GPT-6 Astra · Low: 49.7% $4.95" |
| 12 | Claude Sonnet 5.5 | High | 43.0% | $1.94Cost per attempt | Anthropic | Introducing Claude Sonnet 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5.5 · High: 43.0% $1.94" |
| 13 | Claude Opus 5.5 | Low | 38.5% | $1.29Cost per attempt | Anthropic | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "Opus 5.5 · Low: 38.5% $1.29" |
| 14 | Claude Sonnet 5.5 | Med | 28.8% | $0.83Cost per attempt | Anthropic | Introducing Claude Sonnet 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5.5 · Med: 28.8% $0.83" |
| 15 | Claude Sonnet 5.5 | Low | 20.0% | $0.76Cost per attempt | Anthropic | Introducing Claude Sonnet 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5.5 · Low: 20.0% $0.76" |
official results (Introducing Claude Haiku 5.5)
| Rank | Model | effort | Score | Source cost (metric) | measured/reported by | source |
|---|---|---|---|---|---|---|
| 1 | Claude Sonnet 5.5 | Max | 70.6% | $10.44Cost per attempt | Anthropic | Introducing Claude Haiku 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic); cost series calculated using cache read at $0.10/MTok; the same figure also has a series calculated using cache read at $0.20/MTok, and the cost of that point is $14.15Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5.5 ($0.10 cache reads) · Max: 70.6% $10.44" |
| 2 | Claude Sonnet 5.5 | Xhigh | 61.5% | $4.34Cost per attempt | Anthropic | Introducing Claude Haiku 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic); cost series calculated using cache read at $0.10/MTok; the same figure also has a series calculated using cache read at $0.20/MTok, and the cost of that point is $5.83Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5.5 ($0.10 cache reads) · Xhigh: 61.5% $4.34" |
| 3 | Claude Sonnet 5.5 | High | 43.0% | $1.46Cost per attempt | Anthropic | Introducing Claude Haiku 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic); cost series calculated using cache read at $0.10/MTok; the same figure also has a series calculated using cache read at $0.20/MTok, and the cost of that point is $1.97Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5.5 ($0.10 cache reads) · High: 43.0% $1.46" |
| 4 | Claude Haiku 5.5 | Max | 39.2% | $2.64Cost per attempt | Anthropic | Introducing Claude Haiku 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Haiku 5.5 · Max: 39.2% $2.64" |
| 5 | Claude Haiku 5.5 | Xhigh | 31.5% | $1.75Cost per attempt | Anthropic | Introducing Claude Haiku 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Haiku 5.5 · Xhigh: 31.5% $1.75" |
| 6 | Claude Sonnet 5.5 | Med | 28.8% | $0.68Cost per attempt | Anthropic | Introducing Claude Haiku 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic); cost series calculated using cache read at $0.10/MTok; the same figure also has a series calculated using cache read at $0.20/MTok, and the cost of that point is $0.86Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5.5 ($0.10 cache reads) · Med: 28.8% $0.68" |
| 7 | Claude Haiku 5.5 | High | 24.8% | $1.04Cost per attempt | Anthropic | Introducing Claude Haiku 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Haiku 5.5 · High: 24.8% $1.04" |
| 8 | Claude Haiku 5.5 | Med | 20.3% | $0.68Cost per attempt | Anthropic | Introducing Claude Haiku 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Haiku 5.5 · Med: 20.3% $0.68" |
| 9 | Claude Sonnet 5.5 | Low | 20.0% | $0.62Cost per attempt | Anthropic | Introducing Claude Haiku 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic); cost series calculated using cache read at $0.10/MTok; the same figure also has a series calculated using cache read at $0.20/MTok, and the cost of that point is $0.79Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5.5 ($0.10 cache reads) · Low: 20.0% $0.62" |
| 10 | GPT-6 Luna | not stated | 16.4% | — | not stated | Introducing Claude Haiku 5.5Retrieved 2026-11-12 · Vendor-reported value. table note: see the Haiku 5.5 System Card (effort not stated); values for other vendor models (as listed by Anthropic)Source excerpt16.4%Location Table, Terminal-Bench 4.0 row: 16.4 |
| 11 | Claude Haiku 5.5 | Low | 12.7% | $0.42Cost per attempt | Anthropic | Introducing Claude Haiku 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Haiku 5.5 · Low: 12.7% $0.42" |
official results (Claude Fable 5.2 and Claude Mythos 5.2 System Card)
| Rank | Model | effort | Score | Source cost (metric) | measured/reported by | source |
|---|---|---|---|---|---|---|
| 1 | Claude Fable 5.2 | Max | 67.6% | — | Anthropic | Claude Fable 5.2 and Claude Mythos 5.2 System CardRetrieved 2026-11-14 · Vendor-reported value. Claude Code --bare, max, 66 tasks × 10 trials. safeguards enabled, default fallback (2.5% of requests, 10% of trials). No internet; required resources cached in advance. Standard error ±1.8 pt. Cost not stated.Source excerpt67.6%Location §8.5 “Terminal-Bench 4.0”, body text, Fable 5.2 score |
All model rankings (records)
These rankings are recalculated by comparison group including legacy models and models unavailable in Claude Code / Codex. They differ from the rankings above because the set of models differs.
- View all model rankings (by metric)
- Cost balance (all models, records)
- Overall ranking (all models, records)
Show rankings including legacy and unavailable models (entries: 58)
official results (Accuracy vs Cost figure announced by Anthropic (Opus 5.5 and Sonnet 5.5 pages))
This group combines the same Accuracy vs Cost figure (Terminal-Bench 4.0 and FrontierCode v1.1 main set) on the Claude Opus 5.5 and Claude Sonnet 5.5 announcement page for each model. The scores and costs of all points appearing on both pages match. When the same point appears on both pages, only 1 row is included in the comparison group (the row from the other page is retained as evidence without a key), and the note from the other source is written in the conditions for that 1 row.
| Rank | Model | effort | Score | Source cost (metric) | measured/reported by | source |
|---|---|---|---|---|---|---|
| 1 | Claude Sonnet 5.5 | Max | 70.6% | $12.54Cost per attempt | Anthropic | Introducing Claude Sonnet 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5.5 · Max: 70.6% $12.54" |
| 2 | Claude Opus 5.5 | Xhigh | 66.4% | $7.35Cost per attempt | Anthropic | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "Opus 5.5 · Xhigh: 66.4% $7.35" |
| 3 | Claude Opus 5.5 | Max | 64.8% | $11.24Cost per attempt | Anthropic | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "Opus 5.5 · Max: 64.8% $11.24" |
| 4 | Claude Opus 5.5 | High | 64.2% | $3.88Cost per attempt | Anthropic | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "Opus 5.5 · High: 64.2% $3.88" |
| 5 | Claude Sonnet 5.5 | Xhigh | 61.5% | $5.30Cost per attempt | Anthropic | Introducing Claude Sonnet 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5.5 · Xhigh: 61.5% $5.30" |
| 6 | GPT-6 Astra | High | 57.9% | $7.21Cost per attempt | OpenAI | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor. measured by: the figure makes no distinction / the table footnote says GPT series figures are reported values from OpenAI; the figure's effort label "High" was mapped to high in the OpenAI API reasoning effort; original table footnote: "GPT-6 Astra and GPT-5.6 Sol figures are as reported by OpenAI"Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "GPT-6 Astra · High: 57.9% $7.21" |
| 7 | Claude Opus 5.5 | Med | 57.6% | $2.94Cost per attempt | Anthropic | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "Opus 5.5 · Med: 57.6% $2.94" |
| 7 | GPT-6 Astra | Xhigh | 57.6% | $7.48Cost per attempt | OpenAI | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor. measured by: the figure makes no distinction / the table footnote says GPT series figures are reported values from OpenAI; the figure's effort label "Xhigh" was mapped to xhigh in the OpenAI API reasoning effort; original table footnote: "GPT-6 Astra and GPT-5.6 Sol figures are as reported by OpenAI"Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "GPT-6 Astra · Xhigh: 57.6% $7.48" |
| 9 | GPT-6 Astra | Max | 56.7% | $10.35Cost per attempt | OpenAI | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor. measured by: the figure makes no distinction / the table footnote says GPT series figures are reported values from OpenAI; the figure's effort label "Max" was mapped to max in the OpenAI API reasoning effort; original table footnote: "GPT-6 Astra and GPT-5.6 Sol figures are as reported by OpenAI"Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "GPT-6 Astra · Max: 56.7% $10.35" |
| 10 | Claude Fable 5.1 | Max | 55.8% | $19.50Cost per attempt | Anthropic | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "Fable 5.1 · Max: 55.8% $19.50" |
| 11 | GPT-6 Astra | Med | 53.9% | $6.15Cost per attempt | OpenAI | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor. measured by: the figure makes no distinction / the table footnote says GPT series figures are reported values from OpenAI; the figure's effort label "Med" was mapped to medium in the OpenAI API reasoning effort; original table footnote: "GPT-6 Astra and GPT-5.6 Sol figures are as reported by OpenAI"Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "GPT-6 Astra · Med: 53.9% $6.15" |
| 12 | Claude Opus 5 | Max | 52.3% | $15.83Cost per attempt | Anthropic | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "Opus 5 · Max: 52.3% $15.83" |
| 13 | Claude Fable 5.1 | Xhigh | 51.3% | $15.80Cost per attempt | Anthropic | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "Fable 5.1 · Xhigh: 51.3% $15.80" |
| 14 | Claude Opus 5 | Xhigh | 50.6% | $13.48Cost per attempt | Anthropic | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "Opus 5 · Xhigh: 50.6% $13.48" |
| 15 | GPT-6 Astra | Low | 49.7% | $4.95Cost per attempt | OpenAI | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor. measured by: the figure makes no distinction / the table footnote says GPT series figures are reported values from OpenAI; the figure's effort label "Low" was mapped to low in the OpenAI API reasoning effort; original table footnote: "GPT-6 Astra and GPT-5.6 Sol figures are as reported by OpenAI"Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "GPT-6 Astra · Low: 49.7% $4.95" |
| 16 | Claude Fable 5.1 | High | 49.4% | $10.50Cost per attempt | Anthropic | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "Fable 5.1 · High: 49.4% $10.50" |
| 17 | Claude Opus 5 | High | 47.0% | $10.64Cost per attempt | Anthropic | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "Opus 5 · High: 47.0% $10.64" |
| 18 | Claude Fable 5.1 | Med | 43.4% | $7.80Cost per attempt | Anthropic | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "Fable 5.1 · Med: 43.4% $7.80" |
| 19 | Claude Sonnet 5.5 | High | 43.0% | $1.94Cost per attempt | Anthropic | Introducing Claude Sonnet 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5.5 · High: 43.0% $1.94" |
| 20 | Claude Opus 5 | Med | 41.2% | $7.00Cost per attempt | Anthropic | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "Opus 5 · Med: 41.2% $7" |
| 21 | Claude Fable 5.1 | Low | 40.2% | $5.70Cost per attempt | Anthropic | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "Fable 5.1 · Low: 40.2% $5.70" |
| 22 | Claude Opus 5.5 | Low | 38.5% | $1.29Cost per attempt | Anthropic | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "Opus 5.5 · Low: 38.5% $1.29" |
| 23 | GPT-5.6 Sol | Max | 37.3% | $7.89Cost per attempt | OpenAI | Introducing Claude Sonnet 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor (announced by Anthropic); the figure's effort label "Max" was mapped to max in the OpenAI API reasoning effort; measured by: the Terminal-Bench 4.0 table footnote on the duplicate row for the same point (Opus 5.5 announcement page) states "GPT-6 Astra and GPT-5.6 Sol figures are as reported by OpenAI"Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "GPT-5.6 Sol · Max: 37.3% $7.89" |
| 24 | Claude Sonnet 5.5 | Med | 28.8% | $0.83Cost per attempt | Anthropic | Introducing Claude Sonnet 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5.5 · Med: 28.8% $0.83" |
| 25 | Claude Opus 5 | Low | 28.5% | $4.25Cost per attempt | Anthropic | Introducing Claude Opus 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs Cost: "Opus 5 · Low: 28.5% $4.25" |
| 25 | GPT-5.6 Sol | Xhigh | 28.5% | $5.39Cost per attempt | OpenAI | Introducing Claude Sonnet 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor (announced by Anthropic); the figure's effort label "Xhigh" was mapped to xhigh in the OpenAI API reasoning effort; measured by: the Terminal-Bench 4.0 table footnote on the duplicate row for the same point (Opus 5.5 announcement page) states "GPT-6 Astra and GPT-5.6 Sol figures are as reported by OpenAI"Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "GPT-5.6 Sol · Xhigh: 28.5% $5.39" |
| 27 | GPT-5.6 Sol | High | 26.1% | $4.12Cost per attempt | OpenAI | Introducing Claude Sonnet 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor (announced by Anthropic); the figure's effort label "High" was mapped to high in the OpenAI API reasoning effort; measured by: the Terminal-Bench 4.0 table footnote on the duplicate row for the same point (Opus 5.5 announcement page) states "GPT-6 Astra and GPT-5.6 Sol figures are as reported by OpenAI"Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "GPT-5.6 Sol · High: 26.1% $4.12" |
| 28 | GPT-5.6 Sol | Med | 20.9% | $2.69Cost per attempt | OpenAI | Introducing Claude Sonnet 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor (announced by Anthropic); the figure's effort label "Med" was mapped to medium in the OpenAI API reasoning effort; measured by: the Terminal-Bench 4.0 table footnote on the duplicate row for the same point (Opus 5.5 announcement page) states "GPT-6 Astra and GPT-5.6 Sol figures are as reported by OpenAI"Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "GPT-5.6 Sol · Med: 20.9% $2.69" |
| 29 | Claude Sonnet 5.5 | Low | 20.0% | $0.76Cost per attempt | Anthropic | Introducing Claude Sonnet 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5.5 · Low: 20.0% $0.76" |
| 30 | Claude Sonnet 5 | Max | 10.3% | $11.62Cost per attempt | Anthropic | Introducing Claude Sonnet 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5 · Max: 10.3% $11.62" |
| 31 | GPT-5.6 Sol | Low | 7.9% | $1.46Cost per attempt | OpenAI | Introducing Claude Sonnet 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor (announced by Anthropic); the figure's effort label "Low" was mapped to low in the OpenAI API reasoning effort; measured by: the Terminal-Bench 4.0 table footnote on the duplicate row for the same point (Opus 5.5 announcement page) states "GPT-6 Astra and GPT-5.6 Sol figures are as reported by OpenAI"Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "GPT-5.6 Sol · Low: 7.9% $1.46" |
| 32 | Claude Sonnet 5 | Xhigh | 7.0% | $9.95Cost per attempt | Anthropic | Introducing Claude Sonnet 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5 · Xhigh: 7.0% $9.95" |
| 33 | Claude Sonnet 5 | High | 4.5% | $8.20Cost per attempt | Anthropic | Introducing Claude Sonnet 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5 · High: 4.5% $8.20" |
| 34 | Claude Sonnet 5 | Med | 4.4% | $4.83Cost per attempt | Anthropic | Introducing Claude Sonnet 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5 · Med: 4.4% $4.83" |
| 35 | Claude Sonnet 5 | Low | 3.2% | $3.78Cost per attempt | Anthropic | Introducing Claude Sonnet 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5 · Low: 3.2% $3.78" |
official results (Introducing Claude Haiku 5.5)
| Rank | Model | effort | Score | Source cost (metric) | measured/reported by | source |
|---|---|---|---|---|---|---|
| 1 | Claude Sonnet 5.5 | Max | 70.6% | $10.44Cost per attempt | Anthropic | Introducing Claude Haiku 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic); cost series calculated using cache read at $0.10/MTok; the same figure also has a series calculated using cache read at $0.20/MTok, and the cost of that point is $14.15Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5.5 ($0.10 cache reads) · Max: 70.6% $10.44" |
| 2 | Claude Sonnet 5.5 | Xhigh | 61.5% | $4.34Cost per attempt | Anthropic | Introducing Claude Haiku 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic); cost series calculated using cache read at $0.10/MTok; the same figure also has a series calculated using cache read at $0.20/MTok, and the cost of that point is $5.83Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5.5 ($0.10 cache reads) · Xhigh: 61.5% $4.34" |
| 3 | Claude Sonnet 5.5 | High | 43.0% | $1.46Cost per attempt | Anthropic | Introducing Claude Haiku 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic); cost series calculated using cache read at $0.10/MTok; the same figure also has a series calculated using cache read at $0.20/MTok, and the cost of that point is $1.97Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5.5 ($0.10 cache reads) · High: 43.0% $1.46" |
| 4 | Claude Haiku 5.5 | Max | 39.2% | $2.64Cost per attempt | Anthropic | Introducing Claude Haiku 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Haiku 5.5 · Max: 39.2% $2.64" |
| 5 | Claude Haiku 5.5 | Xhigh | 31.5% | $1.75Cost per attempt | Anthropic | Introducing Claude Haiku 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Haiku 5.5 · Xhigh: 31.5% $1.75" |
| 6 | Claude Sonnet 5.5 | Med | 28.8% | $0.68Cost per attempt | Anthropic | Introducing Claude Haiku 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic); cost series calculated using cache read at $0.10/MTok; the same figure also has a series calculated using cache read at $0.20/MTok, and the cost of that point is $0.86Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5.5 ($0.10 cache reads) · Med: 28.8% $0.68" |
| 7 | Claude Haiku 5.5 | High | 24.8% | $1.04Cost per attempt | Anthropic | Introducing Claude Haiku 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Haiku 5.5 · High: 24.8% $1.04" |
| 8 | Claude Haiku 5.5 | Med | 20.3% | $0.68Cost per attempt | Anthropic | Introducing Claude Haiku 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Haiku 5.5 · Med: 20.3% $0.68" |
| 9 | Claude Sonnet 5.5 | Low | 20.0% | $0.62Cost per attempt | Anthropic | Introducing Claude Haiku 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic); cost series calculated using cache read at $0.10/MTok; the same figure also has a series calculated using cache read at $0.20/MTok, and the cost of that point is $0.79Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Sonnet 5.5 ($0.10 cache reads) · Low: 20.0% $0.62" |
| 10 | GPT-6 Luna | not stated | 16.4% | — | not stated | Introducing Claude Haiku 5.5Retrieved 2026-11-12 · Vendor-reported value. table note: see the Haiku 5.5 System Card (effort not stated); values for other vendor models (as listed by Anthropic)Source excerpt16.4%Location Table, Terminal-Bench 4.0 row: 16.4 |
| 11 | Claude Haiku 5.5 | Low | 12.7% | $0.42Cost per attempt | Anthropic | Introducing Claude Haiku 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Haiku 5.5 · Low: 12.7% $0.42" |
| 12 | Claude Haiku 4.5 | not stated | 0.0% | $0.79Cost per attempt | Anthropic | Introducing Claude Haiku 5.5Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic); no effort because the source effort label "max" is not in the official effort listSource excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Haiku 4.5 · Max: 0.0% $0.79" |
official results (Gemini 4 Carbon (Google DeepMind models page, benchmark table))
| Rank | Model | effort | Score | Source cost (metric) | measured/reported by | source |
|---|---|---|---|---|---|---|
| 1 | Gemini 4 Carbon | High | 65.2% | — | Artificial Analysis | Gemini 4 Carbon (Google DeepMind models page, benchmark table)Retrieved 2026-11-12 · Vendor-reported value. Google published table. methodology PDF: Gemini 4 Carbon/Argon used the highest Gemini API thinking setting, pass@1, and a single trial. Row-specific source: a measured value from Artificial Analysis (labeled Terminal-Bench 4.0 (AA))Source excerptTerminal-Bench 4.0: Gemini 4 Carbon 65.2 |
| 2 | Gemini 4 Argon | High | 57.1% | — | Artificial Analysis | Gemini 4 Carbon (Google DeepMind models page, benchmark table)Retrieved 2026-11-12 · Vendor-reported value. Google published table. methodology PDF: Gemini 4 Carbon/Argon used the highest Gemini API thinking setting, pass@1, and a single trial. Row-specific source: a measured value from Artificial Analysis (labeled Terminal-Bench 4.0 (AA))Source excerptTerminal-Bench 4.0: Gemini 4 Argon 57.1 |
official results (Gemini 4 Argon model card)
| Rank | Model | effort | Score | Source cost (metric) | measured/reported by | source |
|---|---|---|---|---|---|---|
| 1 | Gemini 4 Argon | not stated | 57.4% | — | Gemini 4 Argon model cardRetrieved 2026-11-12 · Vendor-reported value. Self-reported by Google (model card, as of 2026-10). thinking setting not stated. Distinct from the AA measured value of 57.1% on the DeepMind Gemini pageSource excerptTerminal-bench 4.0 | 57.4% |
official results (Gemini 3.8 Flash model card)
| Rank | Model | effort | Score | Source cost (metric) | measured/reported by | source |
|---|---|---|---|---|---|---|
| 1 | Gemini 3.8 Flash | not stated | 19.1% | — | Gemini 3.8 Flash model cardRetrieved 2026-11-12 · Vendor-reported value. Self-reported by Google (model card, as of 2026-09). thinking setting not statedSource excerptTerminal-bench 4.0 | 19.1% |
official results (Grok 4.7 (xAI news))
| Rank | Model | effort | Score | Source cost (metric) | measured/reported by | source |
|---|---|---|---|---|---|---|
| 1 | Grok 4.7 | Xhigh | 37.6% | — | xAI | Grok 4.7 (xAI news)Retrieved 2026-11-12 · Vendor-reported value. Self-reported by xAI. Table heading is Grok 4.7 (xHigh)Source excerptTerminal-Bench 4.0 | Grok 4.7 (xHigh) 37.6 |
| 2 | Grok 4.6 | High | 20.3% | — | xAI | Grok 4.7 (xAI news)Retrieved 2026-11-12 · Vendor-reported value. Self-reported by xAI (comparison column on the Grok 4.7 announcement page)Source excerptTerminal-Bench 4.0 | Grok 4.6 (High) 20.3 |
official results (Introducing Kimi K3.1 Pro)
| Rank | Model | effort | Score | Source cost (metric) | measured/reported by | source |
|---|---|---|---|---|---|---|
| 1 | Kimi K3.1 Pro | Max | 65.7% | — | Moonshot AI | Introducing Kimi K3.1 ProRetrieved 2026-11-12 · Vendor-reported value. Self-reported by Moonshot. footnote: reasoning effort max, temperature 1.0, top-p 0.95. Evaluated with the Claude-Code harnessSource excerptTerminal-Bench 4.0 65.7% |
official results (Introducing Mistral Large 4)
| Rank | Model | effort | Score | Source cost (metric) | measured/reported by | source |
|---|---|---|---|---|---|---|
| 1 | Mistral Large 4 | not stated | 28.3% | — | Mistral AI | Introducing Mistral Large 4Retrieved 2026-11-12 · Vendor-reported value. Self-reported by Mistral (as of the 2026-10-06 preview version; a separate article says the later open weights version improved with additional RL)Source excerptTerminal-Bench 4: 28.3% |
official results (DeepSeek-V4.1-Flash (Hugging Face model card, deepseek-ai))
| Rank | Model | effort | Score | Source cost (metric) | measured/reported by | source |
|---|---|---|---|---|---|---|
| 1 | DeepSeek-V4.1-Flash | not stated | 31.2% | — | DeepSeek | DeepSeek-V4.1-Flash (Hugging Face model card, deepseek-ai)Retrieved 2026-11-13 · Vendor-reported value (self-reported by DeepSeek). Card "Instruct Model": All instruct results below use the maximum effort setting (reasoning_effort=100), temperature=1.0, top_p=0.95. The correspondence of reasoning_effort=100 to the API effort names (low/high/max) has no official statement, so effort is null. Evaluated in Minimal mode of DeepSeek Harness with 1M context (the DeepSeek-V4.1-Flash-1017 card lists the same 31.2 in the Claude Code harness table, so the harness descriptions differ).Source excerptTerminal-Bench 4.0 (Pass@1) | 31.2Location HF model card "Comparison with frontier models (Max reasoning effort)" table, Terminal-Bench 4.0 (Pass@1) row, DS-V4.1-Flash column (rightmost) |
official results (DeepSeek-V4.1-Flash-1017 (Hugging Face model card, deepseek-ai))
| Rank | Model | effort | Score | Source cost (metric) | measured/reported by | source |
|---|---|---|---|---|---|---|
| 1 | DeepSeek-V4.1-Flash-1017 | not stated | 42.3% | — | DeepSeek | DeepSeek-V4.1-Flash-1017 (Hugging Face model card, deepseek-ai)Retrieved 2026-11-13 · Vendor-reported value (self-reported by DeepSeek). Card note: We use the Claude Code harness for Terminal-Bench 4.0 and the Mini-SWE-V2 harness for DeepSWE v1.1. Effort not stated. Claude Code harness.Source excerptTerminal-Bench 4.0 | 42.3Location Comparison table at the start of the HF model card, Terminal-Bench 4.0 row, DS-V4.1-Flash-1017 column (leftmost numeric column) |
official results (Claude Fable 5.2 and Claude Mythos 5.2 System Card)
| Rank | Model | effort | Score | Source cost (metric) | measured/reported by | source |
|---|---|---|---|---|---|---|
| 1 | Claude Fable 5.2 | Max | 67.6% | — | Anthropic | Claude Fable 5.2 and Claude Mythos 5.2 System CardRetrieved 2026-11-14 · Vendor-reported value. Claude Code --bare, max, 66 tasks × 10 trials. safeguards enabled, default fallback (2.5% of requests, 10% of trials). No internet; required resources cached in advance. Standard error ±1.8 pt. Cost not stated.Source excerpt67.6%Location §8.5 “Terminal-Bench 4.0”, body text, Fable 5.2 score |
Values retained as evidence
These are values also published by another source (page) that match values in a combined comparison group. To avoid double-counting, they are excluded from rankings, chart comparison groups, overall aggregations, and role recommendations.
| Model | effort | Score | Source cost (metric) | measured/reported by | source |
|---|---|---|---|---|---|
| GPT-5.6 Sol | Low | 7.9% | $1.46Cost per attempt | OpenAI | Evidence (same value as comparison group Accuracy vs Cost figure announced by Anthropic (Opus 5.5 and Sonnet 5.5 pages))Introducing Claude Opus 5.5official · Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor. measured by: the figure makes no distinction / the table footnote says GPT series figures are reported values from OpenAI; the figure's effort label "Low" was mapped to low in the OpenAI API reasoning effortSource excerptTerminal-Bench 4.0 — Accuracy vs Cost: "GPT-5.6 Sol · Low: 7.9% $1.46" |
| GPT-5.6 Sol | Med | 20.9% | $2.69Cost per attempt | OpenAI | Evidence (same value as comparison group Accuracy vs Cost figure announced by Anthropic (Opus 5.5 and Sonnet 5.5 pages))Introducing Claude Opus 5.5official · Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor. measured by: the figure makes no distinction / the table footnote says GPT series figures are reported values from OpenAI; the figure's effort label "Med" was mapped to medium in the OpenAI API reasoning effortSource excerptTerminal-Bench 4.0 — Accuracy vs Cost: "GPT-5.6 Sol · Med: 20.9% $2.69" |
| GPT-5.6 Sol | High | 26.1% | $4.12Cost per attempt | OpenAI | Evidence (same value as comparison group Accuracy vs Cost figure announced by Anthropic (Opus 5.5 and Sonnet 5.5 pages))Introducing Claude Opus 5.5official · Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor. measured by: the figure makes no distinction / the table footnote says GPT series figures are reported values from OpenAI; the figure's effort label "High" was mapped to high in the OpenAI API reasoning effortSource excerptTerminal-Bench 4.0 — Accuracy vs Cost: "GPT-5.6 Sol · High: 26.1% $4.12" |
| GPT-5.6 Sol | Xhigh | 28.5% | $5.39Cost per attempt | OpenAI | Evidence (same value as comparison group Accuracy vs Cost figure announced by Anthropic (Opus 5.5 and Sonnet 5.5 pages))Introducing Claude Opus 5.5official · Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor. measured by: the figure makes no distinction / the table footnote says GPT series figures are reported values from OpenAI; the figure's effort label "Xhigh" was mapped to xhigh in the OpenAI API reasoning effortSource excerptTerminal-Bench 4.0 — Accuracy vs Cost: "GPT-5.6 Sol · Xhigh: 28.5% $5.39" |
| GPT-5.6 Sol | Max | 37.3% | $7.89Cost per attempt | OpenAI | Evidence (same value as comparison group Accuracy vs Cost figure announced by Anthropic (Opus 5.5 and Sonnet 5.5 pages))Introducing Claude Opus 5.5official · Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs Cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for a model from some other vendor. measured by: the figure makes no distinction / the table footnote says GPT series figures are reported values from OpenAI; the figure's effort label "Max" was mapped to max in the OpenAI API reasoning effortSource excerptTerminal-Bench 4.0 — Accuracy vs Cost: "GPT-5.6 Sol · Max: 37.3% $7.89" |
| Claude Opus 5.5 | Low | 38.5% | $1.29Cost per attempt | Anthropic | Evidence (same value as comparison group Accuracy vs Cost figure announced by Anthropic (Opus 5.5 and Sonnet 5.5 pages))Introducing Claude Sonnet 5.5official · Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Opus 5.5 · Low: 38.5% $1.29" |
| Claude Opus 5.5 | Med | 57.6% | $2.94Cost per attempt | Anthropic | Evidence (same value as comparison group Accuracy vs Cost figure announced by Anthropic (Opus 5.5 and Sonnet 5.5 pages))Introducing Claude Sonnet 5.5official · Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Opus 5.5 · Med: 57.6% $2.94" |
| Claude Opus 5.5 | High | 64.2% | $3.88Cost per attempt | Anthropic | Evidence (same value as comparison group Accuracy vs Cost figure announced by Anthropic (Opus 5.5 and Sonnet 5.5 pages))Introducing Claude Sonnet 5.5official · Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Opus 5.5 · High: 64.2% $3.88" |
| Claude Opus 5.5 | Xhigh | 66.4% | $7.35Cost per attempt | Anthropic | Evidence (same value as comparison group Accuracy vs Cost figure announced by Anthropic (Opus 5.5 and Sonnet 5.5 pages))Introducing Claude Sonnet 5.5official · Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Opus 5.5 · Xhigh: 66.4% $7.35" |
| Claude Opus 5.5 | Max | 64.8% | $11.24Cost per attempt | Anthropic | Evidence (same value as comparison group Accuracy vs Cost figure announced by Anthropic (Opus 5.5 and Sonnet 5.5 pages))Introducing Claude Sonnet 5.5official · Retrieved 2026-11-12 · Vendor-reported value. A point in the figure "Terminal-Bench 4.0 / Accuracy vs. cost" on the Anthropic announcement page. values were taken from each point's aria-label (DOM); the page shows them only in the figure; value for its own model (announced by Anthropic)Source excerptTerminal-Bench 4.0 — Accuracy vs. cost: "Opus 5.5 · Max: 64.8% $11.24" |