BenchLeader

Agnes 2.5 Pro Theta added, Claude Fable 5.2 added (9 Nov)

Agnes 2.5 Pro Theta and Claude Fable 5.2 join the board. 3 new models, 10 top-10 rank changes, 1 price change, 221 new benchmark results. Sources linked.

BenchLeader refreshed 1695 models from 29 of 29 data feeds (benchmark publishers, aggregators and price lists). 3 new models, 10 top-10 rank changes, 1 price change, 221 new or updated benchmark results across 29 data feeds.

New models

  • Agnes 2.5 Pro Theta (Sapiens AI) appeared with 6 benchmark results, entering the BenchLeader Index at #327. (details)
  • Claude Fable 5.2 (Anthropic) appeared with 1 benchmark result. (details)
  • Muse Ember (Meta) is now listed at $0.20/M blended; no independent results yet. (details)

Movement in the top 10

  • Claude Opus 5.5 (max) moved from #1 to #2 in the BenchLeader Index. (details)
  • Gemini 4 Carbon (high) moved from #2 to #5 in the BenchLeader Index. (details)
  • GPT-6 Astra moved from #3 to #6 in the BenchLeader Index. (details)
  • GPT-6 Astra (max) moved from #4 to #7 in the BenchLeader Index. (details)
  • GPT-6.1 Sol (max) moved from #6 to #8 in the BenchLeader Index. (details)
  • Claude Fable 5.1 (max) moved from #5 to #9 in the BenchLeader Index. (details)
  • Claude Fable 5 moved from #9 to #10 in the BenchLeader Index. (details)
  • GPT-6 Astra (high) moved from #7 to #11 in the BenchLeader Index. (details)
  • GPT-6 Astra (xhigh) moved from #8 to #13 in the BenchLeader Index. (details)
  • Claude Fable 5.1 moved from #10 to #14 in the BenchLeader Index. (details)

Price changes

  • Muse Glimmer (high) cost per task moved from $0.16/M to $0.17/M (5% up). (details)

New benchmark results

  • Arena Text: 1 new result, including Inkling 2 Small (board)
  • Arena Hard Prompts: 1 new result, including Inkling 2 Small (board)
  • Arena Coding: 1 new result, including Inkling 2 Small (board)
  • Arena Agent: 1 new result, including Inkling 2 Small (board)
  • Arena Creative Writing: 1 new result, including Inkling 2 Small (board)
  • Arena Instruction Following: 1 new result, including Inkling 2 Small (board)
  • Arena Multi-turn: 1 new result, including Inkling 2 Small (board)
  • Arena Longer Queries: 1 new result, including Inkling 2 Small (board)
  • CritPt v2 hard subset (AA): 1 new result, including gpt-oss-120b (board)

Score revisions

  • Claude Opus 5.5 (max) on AA-3D: 1196 → 1202. (details)
  • Claude Opus 5.5 (max) on Terminal-Bench: 64.9% → 65.8%. (details)
  • GPT-6 Astra on LiveBench 2.0: 85.6% → 84.5%. (details)
  • GPT-6 Astra on LiveBench Long Agentic: 77.5% → 75.6%. (details)
  • GPT-6 Astra on LiveBench 2.0 Coding: 93.7% → 87.5%. (details)
  • GPT-6 Astra on LiveBench Research: 85.2% → 90.6%. (details)
  • GPT-6 Astra on LiveBench Design: 82% → 80%. (details)
  • GPT-6 Astra (max) on Arena Agent: 14.7 → 14.2. (details)
  • GPT-6 Astra (max) on Terminal-Bench: 58.2% → 60.3%. (details)
  • GPT-6.1 Sol (max) on Arena Maths: 1504 → 1498. (details)
  • GPT-6.1 Sol (max) on Terminal-Bench: 58.2% → 60.9%. (details)
  • Claude Fable 5.1 (max) on Terminal-Bench: 57.9% → 57.3%. (details)
  • GPT-6 Astra (high) on Terminal-Bench: 57.9% → 60%. (details)
  • GPT-6 Astra (xhigh) on Terminal-Bench: 57.9% → 59.7%. (details)
  • Claude Fable 5.1 on LiveBench 2.0: 85.7% → 84%. (details)
  • Claude Fable 5.1 on LiveBench Long Agentic: 77.9% → 74.2%. (details)
  • Claude Fable 5.1 on LiveBench 2.0 Coding: 97.7% → 89.7%. (details)
  • Claude Fable 5.1 on LiveBench Research: 82% → 88.5%. (details)
  • Claude Fable 5.1 on LiveBench Design: 83.4% → 81.6%. (details)
  • Claude Fable 5.1 on Tax Agent Bench: 49.2% → 50.8%. (details)
  • …and 192 more.

Speed changes

  • GPT-6 Astra time to first token changed from 7.98 s to 4.75 s. (details)
  • Claude Fable 5 output speed changed from 29 tok/s to 55 tok/s. (details)
  • GPT-5.6 Sol time to first token changed from 1.99 s to 3.29 s. (details)
  • GPT-6.1 Terra time to first token changed from 1.84 s to 2.56 s. (details)
  • GPT-5.4 Pro time to first token changed from 6.24 s to 8.94 s. (details)
  • GPT-5.6 Terra time to first answer changed from 28.48 s to 17.43 s. (details)
  • GLM-5 time to first answer changed from 56.97 s to 27.97 s. (details)
  • GLM-5 thinking time changed from 54.16 s to 25.06 s. (details)
  • Claude Opus 4.5 time to first answer changed from 23.63 s to 12.63 s. (details)
  • DeepSeek V4 Pro time to first token changed from 2.40 s to 1.47 s. (details)
  • GLM 5.3 time to first token changed from 0.32 s to 0.52 s. (details)
  • DeepSeek V4 Flash time to first token changed from 1.37 s to 0.87 s. (details)
  • MiMo-V2.5 output speed changed from 75 tok/s to 35 tok/s. (details)
  • MiMo-V2.5 time to first token changed from 1.21 s to 2.78 s. (details)
  • DeepSeek V3.2 time to first token changed from 1.30 s to 1.98 s. (details)
  • …and 85 more.

Today's top five

  1. Claude Fable 5.2 (max) — 72.7
  2. Gemini 4 Carbon (high) — 68.9
  3. Claude Opus 5.5 (max) — 68.4
  4. GPT-6 Astra (max) — 67.3
  5. Claude Fable 5.1 (max) — 66.7

More: New AI model releases this week, and the full LLM leaderboard 2026.

This digest is generated automatically from the day's data changes. Every line links to the page where you can check the numbers and their source.

Head to head

Get this in your inbox

No ads, no tracking pixels, unsubscribe in one click.
What to receive