The AI models leading today, and the ones we build on.

Ranked by the Epoch Capabilities Index, from Epoch’s benchmark data dated 11 November 2026. Below the table: which model we would pick for which job, and the numbers behind each pick.

The top ten

Top 10 by the Epoch Capabilities Index.

Model rankings by Epoch Capabilities Index
ModelDeveloperCapabilities IndexGPQA DiamondSimpleQA VerifiedWe build on
Claude Fable 5.2AnthropicAnthropic173.4493.9%94.0%Amtec builds on
Gemini 4 CarbonGoogle DeepMindGoogle DeepMind169.7696.5%77.7%
GPT-6 AstraOpenAIOpenAI169.0095.8%75.6%Amtec builds on
Claude Opus 5.5AnthropicAnthropic168.5490.6%72.2%Amtec builds on
GPT-6.1 SolOpenAIOpenAI168.4895.4%73.9%Amtec builds on
GPT-6.1 TerraOpenAIOpenAI167.6794.3%57.3%Amtec builds on
Claude Fable 5.1AnthropicAnthropic166.99Not tested70.8%Amtec builds on
Claude Sonnet 5.5AnthropicAnthropic166.5695.6%46.5%Amtec builds on
Gemini 4 ArgonGoogle DeepMindGoogle DeepMind166.4896.4%76.0%
Claude Opus 5AnthropicAnthropic164.6893.9%59.9%Amtec builds on

Capabilities Index: overall performance across several tests. GPQA Diamond: expert-level science questions. SimpleQA Verified: factual accuracy.

Source: Epoch AI, Benchmarking data (CC BY 4.0). Data as of 11 November 2026.

Which model for which job

The index ranks how capable a model is. It does not say what a model costs to do your work, or which work it does best. Epoch tracks that too, so here is how we read it.

Which models: the ten in the table above, plus Claude Haiku 5.5, the smallest model we build on. Every figure comes from Epoch’s benchmark data dated 11 November 2026. Each is an average over one test’s tasks, not a promise for any one job.

Professional work is close at the top

APEX-Agents gives an agent the multi-step tasks that banking, consulting and law teams hand out, and checks the finished work. Each bar is a model’s best setting. Bars less than about 12 points apart can be inside the margin of error.

  1. Gemini 4 Carbon84.2%
  2. Gemini 4 Argon82.4%
  3. Claude Fable 5.280.1%
  4. Claude Sonnet 5.575.5%
  5. Claude Opus 5.573.5%
  6. Claude Fable 5.168.6%
  7. GPT-6.1 Terra68.3%
  8. Claude Opus 565.8%
  9. GPT-6 Astra64.7%
  10. GPT-6.1 Sol60%
  11. Claude Haiku 5.556.1%

Coding: what each extra point costs

CursorBench sets coding tasks in real codebases and records what each attempt costs. Each line is one model, from its lowest effort setting to its highest. Up and to the left is better: more solved for less. It measures coding only, so it does not rank models for other work.

Best value30%40%50%60%$0.10$0.25$0.50$1$2$5$10$20$50Cost per task (log scale)Claude Opus 5.5, low effort: 43.7% solved, $1.17 a taskClaude Opus 5.5, medium effort: 52.5% solved, $2.91 a taskClaude Opus 5.5, high effort: 56% solved, $3.97 a taskClaude Opus 5.5, extra high effort: 56% solved, $6.98 a taskClaude Opus 5.5, max effort: 57.8% solved, $13.43 a taskOpus 5.5Claude Sonnet 5.5, low effort: 35.8% solved, $0.37 a taskClaude Sonnet 5.5, medium effort: 39.2% solved, $0.52 a taskClaude Sonnet 5.5, high effort: 47.8% solved, $1.20 a taskClaude Sonnet 5.5, extra high effort: 53.1% solved, $2.81 a taskClaude Sonnet 5.5, max effort: 55.5% solved, $7.05 a taskSonnet 5.5Claude Haiku 5.5, low effort: 30.9% solved, $0.08 a taskClaude Haiku 5.5, medium effort: 36.9% solved, $0.17 a taskClaude Haiku 5.5, high effort: 42.3% solved, $0.32 a taskClaude Haiku 5.5, extra high effort: 44.3% solved, $0.56 a taskClaude Haiku 5.5, max effort: 48.4% solved, $1.12 a taskHaiku 5.5Claude Fable 5.2, low effort: 53.2% solved, $3.94 a taskClaude Fable 5.2, medium effort: 54.8% solved, $4.66 a taskClaude Fable 5.2, high effort: 56.9% solved, $9.70 a taskClaude Fable 5.2, extra high effort: 56.3% solved, $15.48 a taskClaude Fable 5.2, max effort: 60% solved, $32.79 a taskFable 5.2Gemini 4 Carbon, low effort: 58.6% solved, $11.89 a taskGemini 4 Carbon, medium effort: 60.1% solved, $8.49 a taskGemini 4 Carbon, high effort: 60.4% solved, $9.22 a taskGemini 4 CarbonGemini 4 Argon, minimal effort: 48.8% solved, $3.54 a taskGemini 4 Argon, low effort: 50.6% solved, $4.09 a taskGemini 4 Argon, medium effort: 52.8% solved, $4.39 a taskGemini 4 Argon, high effort: 52.9% solved, $4.57 a taskGemini 4 ArgonClaude Opus 5, low effort: 40.7% solved, $4.87 a taskClaude Opus 5, medium effort: 43.3% solved, $6.94 a taskClaude Opus 5, high effort: 44.7% solved, $9.00 a taskClaude Opus 5, extra high effort: 46.1% solved, $11.43 a taskClaude Opus 5, max effort: 46.6% solved, $11.95 a taskOpus 5Claude Fable 5.1, low effort: 45.1% solved, $5.44 a taskClaude Fable 5.1, medium effort: 46.8% solved, $7.05 a taskClaude Fable 5.1, high effort: 49.2% solved, $9.08 a taskClaude Fable 5.1, extra high effort: 51.6% solved, $13.01 a taskClaude Fable 5.1, max effort: 51.8% solved, $17.28 a taskFable 5.1Best value30%40%50%60%$0.10$1$10$50Cost per task (log scale)Claude Opus 5.5, low effort: 43.7% solved, $1.17 a taskClaude Opus 5.5, medium effort: 52.5% solved, $2.91 a taskClaude Opus 5.5, high effort: 56% solved, $3.97 a taskClaude Opus 5.5, extra high effort: 56% solved, $6.98 a taskClaude Opus 5.5, max effort: 57.8% solved, $13.43 a taskOpus 5.5Claude Sonnet 5.5, low effort: 35.8% solved, $0.37 a taskClaude Sonnet 5.5, medium effort: 39.2% solved, $0.52 a taskClaude Sonnet 5.5, high effort: 47.8% solved, $1.20 a taskClaude Sonnet 5.5, extra high effort: 53.1% solved, $2.81 a taskClaude Sonnet 5.5, max effort: 55.5% solved, $7.05 a taskSonnet 5.5Claude Haiku 5.5, low effort: 30.9% solved, $0.08 a taskClaude Haiku 5.5, medium effort: 36.9% solved, $0.17 a taskClaude Haiku 5.5, high effort: 42.3% solved, $0.32 a taskClaude Haiku 5.5, extra high effort: 44.3% solved, $0.56 a taskClaude Haiku 5.5, max effort: 48.4% solved, $1.12 a taskHaiku 5.5Claude Fable 5.2, low effort: 53.2% solved, $3.94 a taskClaude Fable 5.2, medium effort: 54.8% solved, $4.66 a taskClaude Fable 5.2, high effort: 56.9% solved, $9.70 a taskClaude Fable 5.2, extra high effort: 56.3% solved, $15.48 a taskClaude Fable 5.2, max effort: 60% solved, $32.79 a taskFable 5.2Gemini 4 Carbon, low effort: 58.6% solved, $11.89 a taskGemini 4 Carbon, medium effort: 60.1% solved, $8.49 a taskGemini 4 Carbon, high effort: 60.4% solved, $9.22 a taskGemini 4 CarbonGemini 4 Argon, minimal effort: 48.8% solved, $3.54 a taskGemini 4 Argon, low effort: 50.6% solved, $4.09 a taskGemini 4 Argon, medium effort: 52.8% solved, $4.39 a taskGemini 4 Argon, high effort: 52.9% solved, $4.57 a taskGemini 4 ArgonClaude Opus 5, low effort: 40.7% solved, $4.87 a taskClaude Opus 5, medium effort: 43.3% solved, $6.94 a taskClaude Opus 5, high effort: 44.7% solved, $9.00 a taskClaude Opus 5, extra high effort: 46.1% solved, $11.43 a taskClaude Opus 5, max effort: 46.6% solved, $11.95 a taskOpus 5Claude Fable 5.1, low effort: 45.1% solved, $5.44 a taskClaude Fable 5.1, medium effort: 46.8% solved, $7.05 a taskClaude Fable 5.1, high effort: 49.2% solved, $9.08 a taskClaude Fable 5.1, extra high effort: 51.6% solved, $13.01 a taskClaude Fable 5.1, max effort: 51.8% solved, $17.28 a taskFable 5.1
Claude Opus 5.5Claude Sonnet 5.5Claude Haiku 5.5Claude Fable 5.2Gemini 4 CarbonGemini 4 ArgonClaude Opus 5Claude Fable 5.1

No CursorBench result yet for GPT-6 Astra or GPT-6.1 Sol or GPT-6.1 Terra, so they are not on the chart.

What we would pick

  • Coding in an existing codebase

    Claude Opus 5.5, high effort

    On CursorBench it solves 56% of tasks at $3.97 a task. Every setting that solves more costs at least 2.1 times as much per task: the cheapest is Gemini 4 Carbon at medium effort, 60.1% for $8.49. Max effort adds 1.8 points for 3.4 times the cost.

  • Routine work at volume

    Claude Sonnet 5.5, high effort

    85% of Opus 5.5's high-effort CursorBench score for 30% of the cost: $1.20 a task. At max effort it passes 75.5% of the APEX-Agents professional tasks.

  • Sorting an email, routing a request or pulling one field

    Claude Haiku 5.5

    At high effort it solves 42.3% of CursorBench tasks for $0.32 a task, where Opus 5.5 at high effort costs 12 times as much. At max effort it passes 56.1% of the APEX-Agents tasks. For a narrow decision we check it on the client’s own examples before relying on it.

  • Document-heavy professional work

    Worth a look: Gemini 4 Carbon

    The top APEX-Agents score, 84.2%. Gemini 4 Argon, Claude Fable 5.2, Claude Sonnet 5.5 and Claude Opus 5.5 are within the margin of error, so we compare them on the client’s own documents before choosing.

  • Where extra money buys no more on this test

    Claude Opus 5 and Fable 5.1

    On CursorBench, Opus 5.5 at high effort scores higher than every setting of both and costs less per task than any of them.

Table view
Coding: CursorBench by effort setting
ModelEffortSolvedCost per task
Claude Haiku 5.5low30.9%$0.08
Claude Haiku 5.5medium36.9%$0.17
Claude Haiku 5.5high42.3%$0.32
Claude Sonnet 5.5low35.8%$0.37
Claude Sonnet 5.5medium39.2%$0.52
Claude Haiku 5.5extra high44.3%$0.56
Claude Haiku 5.5max48.4%$1.12
Claude Opus 5.5low43.7%$1.17
Claude Sonnet 5.5high47.8%$1.20
Claude Sonnet 5.5extra high53.1%$2.81
Claude Opus 5.5medium52.5%$2.91
Gemini 4 Argonminimal48.8%$3.54
Claude Fable 5.2low53.2%$3.94
Claude Opus 5.5high56%$3.97
Gemini 4 Argonlow50.6%$4.09
Gemini 4 Argonmedium52.8%$4.39
Gemini 4 Argonhigh52.9%$4.57
Claude Fable 5.2medium54.8%$4.66
Claude Opus 5low40.7%$4.87
Claude Fable 5.1low45.1%$5.44
Claude Opus 5medium43.3%$6.94
Claude Opus 5.5extra high56%$6.98
Claude Fable 5.1medium46.8%$7.05
Claude Sonnet 5.5max55.5%$7.05
Gemini 4 Carbonmedium60.1%$8.49
Claude Opus 5high44.7%$9.00
Claude Fable 5.1high49.2%$9.08
Gemini 4 Carbonhigh60.4%$9.22
Claude Fable 5.2high56.9%$9.70
Claude Opus 5extra high46.1%$11.43
Gemini 4 Carbonlow58.6%$11.89
Claude Opus 5max46.6%$11.95
Claude Fable 5.1extra high51.6%$13.01
Claude Opus 5.5max57.8%$13.43
Claude Fable 5.2extra high56.3%$15.48
Claude Fable 5.1max51.8%$17.28
Claude Fable 5.2max60%$32.79
Professional work: APEX-Agents, best setting
ModelPassed first try
Gemini 4 Carbon84.2%
Gemini 4 Argon82.4%
Claude Fable 5.280.1%
Claude Sonnet 5.575.5%
Claude Opus 5.573.5%
Claude Fable 5.168.6%
GPT-6.1 Terra68.3%
Claude Opus 565.8%
GPT-6 Astra64.7%
GPT-6.1 Sol60%
Claude Haiku 5.556.1%

Data: Epoch AI, Benchmarking data (CC BY 4.0), archive files dated 11 November 2026. Cost per task is Epoch’s figure for one attempt.

Find the work an agent could take off your team.

Find your bottlenecks