Breaking Tanker hits docked cruise ship on Mississippi River in New Orleans 10 sources · since 2:41 PM EST (5 hr ago)
How it works

WhatsNewPaper

Vol. 3 · No. 317 Every side. Side by side. Friday, November 13, 2026

AI Scoreboard

When a lab ships a new model, this is how it measures up: independent test scores, price and release date, next to the models it competes with.

Just released

AnthropicNew

Claude Fable 5.2

  • Released Nov. 9, 2026
  • Paid API
  • $8 in / $40 out per million tokens
  • Reads up to 1M tokens at once

1st of 44 overall on Epoch AI’s index, scoring 173.4 (likely between 169.9 and 177.5).

TestClaude Fable 5.2Gemini 4 CarbonGoogle DeepMindGPT-6 AstraOpenAIMuse WonderMeta AIClaude Opus 5.5Same lab, earlier
OverallEpoch Capabilities Index173.4 (best here)169.8169.0161.7168.5
ScienceGPQA Diamond96%96.5% (best here)95.8%—Not tested yet94.1%
MathFrontierMath, Tiers 1–395.1% (best here)90.5%93.7%82.1%91.2%
PuzzlesARC-AGI-293.8%94.2%95% (best here)—Not tested yet93.3%
Science codeSciCode67.7% (best here)64.2%56.5%60.3%66.9%
Office workAPEX-Agents80.1%84.2% (best here)64.7%66.1%73.5%
FactsSimpleQA Verified94% (best here)77.7%75.6%—Not tested yet72.2%
PhysicsCritPt35.7%36% (best here)31.7%28.3%31.7%
Priceper million tokens$8 in / $40 out$2 in / $10 out$10 in / $50 out$4 in / $20 out$4 in / $20 out

No published result yet for Claude Fable 5.2 on: Coding.

Thinking Machines

Inkling 2 Small

  • Released Nov. 5, 2026
  • $0.30 in / $1 out per million tokens
  • Reads up to 1M tokens at once

Epoch AI has not given it an overall score yet. The test results so far:

TestInkling 2 SmallClaude Fable 5.2AnthropicGemini 4 CarbonGoogle DeepMindGPT-6 AstraOpenAI
OverallEpoch Capabilities Index—Not tested yet173.4 (best here)169.8169.0
PuzzlesARC-AGI-281.7%93.8%94.2%95% (best here)
Science codeSciCode52.3%67.7% (best here)64.2%56.5%
Office workAPEX-Agents46.8%80.1%84.2% (best here)64.7%
Priceper million tokens$0.30 in / $1 out$8 in / $40 out$2 in / $10 out$10 in / $50 out

No published result yet for Inkling 2 Small on: Science, Math, Coding, Facts, Physics.

ByteDance

Doubao Seed Evolving (Nov 2026)

  • Released Nov. 2, 2026
  • Paid API

Epoch AI has not given it an overall score yet. The test results so far:

TestDoubao Seed Evolving (Nov 2026)Claude Fable 5.2AnthropicGemini 4 CarbonGoogle DeepMindGPT-6 AstraOpenAI
OverallEpoch Capabilities Index—Not tested yet173.4 (best here)169.8169.0
ScienceGPQA Diamond93.1%96%96.5% (best here)95.8%
FactsSimpleQA Verified41.9%94% (best here)77.7%75.6%
Priceper million tokens—$8 in / $40 out$2 in / $10 out$10 in / $50 out

No published result yet for Doubao Seed Evolving (Nov 2026) on: Math, Puzzles, Coding, Science code, Office work, Physics.

GoogleNew

Gemini 4 Hydrogen

  • Released Nov. 12, 2026
  • $1.50 in / $7.50 out per million tokens
  • Reads up to 1M tokens at once

No independent scores yet. Testers usually publish within days of a release.

See the field it joins →

Our coverage →

Google

Gemini Omni 1.2 Flash

  • Released Nov. 5, 2026
  • $1.50 in / $9 out per million tokens
  • Reads up to 131K tokens at once

No independent scores yet. Testers usually publish within days of a release.

See the field it joins →

Google

Gemini Omni 1.1 Flash

  • Released Nov. 4, 2026
  • $1.50 in / $9 out per million tokens
  • Reads up to 131K tokens at once

No independent scores yet. Testers usually publish within days of a release.

See the field it joins →

Tencent

Hy4

  • Released Nov. 4, 2026
  • $0.42 in / $1.25 out per million tokens
  • Reads up to 1M tokens at once

No independent scores yet. Testers usually publish within days of a release.

See the field it joins →

Also new: Mistral Large 4 (Mistral AI, Oct. 29) · Gemini 4 Carbon (Google DeepMind, Oct. 28) · GPT-6.1 Terra (OpenAI, Oct. 26) · Kimi K3.1 Pro (Moonshot, Oct. 23) · DeepSeek V4.1 Flash 1017 (DeepSeek, Oct. 17)

Every model, test by test

50 models. Tap a column to sort by it.

Model
Inkling 2 SmallThinking Machines Not ranked yet——81.7%—52.3%46.8%—— $0.30 / $1 Nov. 5
Doubao Seed Evolving (Nov 2026)ByteDance Not ranked yet93.1%—————41.9%— — Nov. 2
Claude Fable 5.2NewAnthropic 173.496%95.1%93.8%—67.7%80.1%94%35.7% $8 / $40 Nov. 9
Gemini 4 CarbonGoogle DeepMind 169.896.5%90.5%94.2%—64.2%84.2%77.7%36% $2 / $10 Oct. 28
GPT-6 AstraOpenAI 169.095.8%93.7%95%53.3%56.5%64.7%75.6%31.7% $10 / $50 Sept. 3
GPT-6.1 SolOpenAI 168.595.4%93.7%94.2%50.2%55.8%60%73.9%31.7% $2 / $10 Sept. 29
Claude Opus 5.5Anthropic 168.594.1%91.2%93.3%54.6%66.9%73.5%72.2%31.7% $4 / $20 Sept. 22
GPT-6.1 TerraOpenAI 167.794.3%90.9%——55.1%68.3%57.3%28% $1 / $5 Oct. 26
Claude Fable 5.1Anthropic 167.0—90.2%90%50.9%63.1%68.6%70.8%31.1% $10 / $50 Sept. 1
Claude Sonnet 5.5Anthropic 166.695.6%88.8%83.9%52.1%61%75.5%46.5%31.4% $2 / $10 Sept. 28
Gemini 4 ArgonGoogle DeepMind 166.596.4%86%91.3%47.6%61.8%82.4%76%27.1% $2 / $10 Sept. 30
Claude Opus 5Anthropic 164.793.9%85.6%90.4%53.4%56.4%65.8%59.9%29.1% $5 / $25 July 24
Claude Fable 5Anthropic 164.193.9%87%89.2%53.5%61%63.6%70.7%28.6% $10 / $50 June 9
GPT-6 SolOpenAI 163.894.3%89.8%89.6%49.3%57.6%54.3%60.7%30.9% $2 / $10 Sept. 22
GPT-5.6 SolOpenAI 163.093.5%89.1%92.5%47.5%57.8%51.4%69.7%32.3% $4 / $20 July 9
GPT-5.5 ProOpenAI 162.793.9%87.7%84.6%————30.6% $30 / $180 April 23
Muse WonderMeta AI 161.7—82.1%——60.3%66.1%—28.3% $4 / $20 Oct. 12
Kimi K3.1 ProMoonshot · open weights 161.494.2%78.2%——60.8%65.5%62.2%24.6% $2 / $10 Oct. 23
GPT-5.6 TerraOpenAI 160.793.3%86%83.9%41.3%55%58.2%43.2%30% $2 / $12 July 9
Claude Haiku 5.5Anthropic 160.489.6%75.1%61.3%46.4%55%56.1%23.8%22.6% $0.10 / $0.50 Oct. 7
GPT-5.4 ProOpenAI 160.294.6%82.5%83.3%———46.3%30% $30 / $180 March 5
GPT-5.5OpenAI 160.194%85.3%85%43%56.1%55.1%63%27.1% $5 / $30 April 23
Claude Opus 4.8Anthropic 159.991%80%72.1%46.5%53.5%48.9%53%20.9% $5 / $25 May 28
DeepSeek V4.1 Flash 1017DeepSeek · open weights 159.493.2%71.2%78.1%—55.9%69.8%42.5%24% $0.04 / $0.50 Oct. 17
Gemini 3.8 FlashGoogle DeepMind 158.895.4%68.4%89.2%41.2%56.6%64.3%69.7%18.3% $0.83 / $4.13 Sept. 2
Kimi K3Moonshot · open weights 158.693.1%72.2%60.4%44.2%59.5%47.8%50.6%23.4% $0.19 / $16 July 16
Gemini 3.7 FlashGoogle DeepMind 158.494.8%71.6%84.6%43.6%59.8%67.8%69.2%14.3% $0.75 / $3.75 Aug. 13
Muse Spark 1.3Meta AI 158.2—74.4%——59.7%58.6%—26% $1.25 / $4.25 Sept. 2
GPT-5.4OpenAI 157.993.3%78.6%74%—56.6%52.4%45.1%23.4% $2.50 / $15 March 5
Grok 4.6xAI 157.894%66%67.1%48%56.5%65.3%49.3%19.7% $2 / $6 Aug. 12
Qwen 3.8 MaxAlibaba 157.692.7%74.7%——53.2%63.3%45.8%20% — Aug. 2
Claude Opus 4.7Anthropic 157.490.2%70.2%75.8%38.5%54.5%49.2%51.7%12% $5 / $25 April 16
GPT-5.2 ProOpenAI 157.3—74%54.2%————— $21 / $168 Dec. 11
GLM-5.3Z.ai (Zhipu AI) · open weights 157.190.9%68.8%—40.1%59%56.6%41%19.1% $0.08 / $3.60 Aug. 14
Claude Sonnet 5Anthropic 156.990.5%65.6%—42.7%54.3%54.5%33.7%16.9% $2 / $10 June 30
DeepSeek V4.1 FlashDeepSeek · open weights 156.789.8%67.4%72.9%43.9%51.9%39.5%42.9%14.3% $0.05 / $0.20 Sept. 9
GPT-5.6 LunaOpenAI 156.791.6%82.1%59.5%39.8%53.6%43%41%20.6% $0.20 / $1.20 July 9
GPT-6 LunaOpenAI 156.490.5%78.9%59.3%42.4%54.6%44.3%41.4%19.4% $0.10 / $0.50 Sept. 22
Claude Opus 4.6Anthropic 156.390.5%66%69.2%26.6%—46.3%47%— $5 / $25 Feb. 5
Grok 4.7xAI 156.292.7%53%61.4%47.6%57.8%54.6%56%18% $2 / $6 Sept. 21
Gemini 3.1 ProGoogle DeepMind 156.294.4%59.6%77.1%—58.7%35.1%73.5%17.7% — Feb. 19
Qwen3.8 Max (0902)Alibaba 156.092.3%65.6%——52.1%—47.3%17.7% $2 / $6 Sept. 1
Muse Spark OpenMeta AI · open weights 155.488.6%69.8%57.5%—58.4%33.4%56.6%14.9% $1.25 / $4.25 Oct. 15
Mistral Large 4Mistral AI · open weights 154.690.2%—————22.4%11.1% $0.70 / $2.10 Oct. 29
Mistral Large 4 PreviewMistral AI · open weights 152.788.1%58.6%—39.4%54.2%51.6%20%10.6% — Oct. 6
Beam 1.0Reflection · open weights 150.9——23.5%—48.5%48.8%—13.1% — Oct. 5
Gemini 4 HydrogenNewGoogle ————————— $1.50 / $7.50 Nov. 12
Gemini Omni 1.1 FlashGoogle ————————— $1.50 / $9 Nov. 4
Gemini Omni 1.2 FlashGoogle ————————— $1.50 / $9 Nov. 5
Hy4Tencent ————————— $0.42 / $1.25 Nov. 4

What the tests measure

Overall Epoch Capabilities Index · Epoch AI
One overall score. Epoch AI combines a model’s results on dozens of tests into a single number, so models that took different tests can still be compared. Higher is better; a few points apart is within the margin of error.
Science GPQA Diamond · run by Epoch AI
Graduate-level multiple-choice questions in physics, chemistry and biology. Guessing scores 25%.
Math FrontierMath, Tiers 1–3 · run by Epoch AI
Original math problems, from hard undergraduate level to early research level, kept private so a model cannot have seen the answers.
Puzzles ARC-AGI-2 · ARC Prize Foundation
Visual pattern puzzles that most people can solve. Each one needs working out a new rule, not recalling facts.
Coding FrontierCode · Cognition
Hard programming tasks an AI agent has to finish on its own.
Science code SciCode · Artificial Analysis
Writing code to solve problems taken from real scientific research.
Office work APEX-Agents · Mercor
Long professional tasks of the kind done in banking, consulting and law, with one attempt each.
Facts SimpleQA Verified · run by Epoch AI
Short factual questions with one right answer, answered without looking anything up.
Physics CritPt · CritPt
Unpublished research-level physics problems.

A test score is not the whole model. Labs tune for well-known tests, small gaps are often noise, and none of these measures whether a model is right for a given job. A dash means nobody has published that test for that model yet.

Scores: Epoch AI Benchmarking Hub (CC BY 4.0), each the model’s best result across the settings tested, as of Nov. 14, 2026. Release dates and prices: OpenRouter. The paper runs none of these tests.