GPT-6.2 Astra up to #2 (19 Nov 2026)
GPT-6.2 Astra moved #3 → #2; Llama 3.2 Instruct 11B (Vision) got cheaper. 6 top-10 rank changes, 61 price changes, 380 new benchmark results.
BenchLeader refreshed 1776 models from 29 of 29 data feeds (benchmark publishers, aggregators and price lists). 6 top-10 rank changes, 61 price changes, 380 new or updated benchmark results across 29 data feeds.
Movement in the top 10
- GPT-6.2 Astra (max) moved from #3 to #2 in the BenchLeader Index. (details)
- Claude Fable 5.2 (xhigh) moved from #2 to #3 in the BenchLeader Index. (details)
- Claude Fable 5.2 (high) moved from #6 to #7 in the BenchLeader Index. (details)
- GPT-6 Astra (max) moved from #7 to #9 in the BenchLeader Index. (details)
- Claude Fable 5.1 (max) moved from #9 to #10 in the BenchLeader Index. (details)
- GPT-6 Astra moved from #10 to #11 in the BenchLeader Index. (details)
Price changes
- Gemini 4 Carbon (high) cost per task moved from $4.90/M to $7.76/M (58% up). (details)
- Gemini 4 Carbon (medium) cost per task moved from $4.16/M to $6.59/M (58% up). (details)
- Gemini 4 Carbon (low) cost per task moved from $4.68/M to $7.19/M (53% up). (details)
- DeepSeek V3.1 Terminus (thinking) input price moved from $0.27/M to $1.64/M (506% up). (details)
- DeepSeek V3.1 Terminus (thinking) output price moved from $1.00/M to $2.75/M (175% up). (details)
- Nemotron 3 Ultra 550B A55B input price moved from $0.60/M to $0.55/M (8% down). (details)
- Nemotron 3 Ultra 550B A55B cached input price moved from $0.12/M to $0.11/M (8% down). (details)
- Hy3 output price moved from $0.55/M to $0.58/M (5% up). (details)
- GLM 5.1 input price moved from $1.19/M to $1.28/M (8% up). (details)
- GLM 5.1 output price moved from $3.74/M to $4.07/M (9% up). (details)
- GLM 5.1 cached input price moved from $0.60/M to $0.26/M (57% down). (details)
- Beam 1.0 501B A23B input price moved from $0.30/M to $0.25/M (17% down). (details)
- Beam 1.0 501B A23B cached input price moved from $0.040/M to $0.050/M (25% up). (details)
- DeepSeek V3.1 Terminus input price moved from $0.27/M to $1.64/M (506% up). (details)
- DeepSeek V3.1 Terminus output price moved from $1.00/M to $2.75/M (175% up). (details)
- Nemotron 3 Ultra 550B A55B (thinking) input price moved from $0.60/M to $0.55/M (8% down). (details)
- Nemotron 3 Ultra 550B A55B (thinking) cached input price moved from $0.12/M to $0.11/M (8% down). (details)
- DeepSeek V3 0324 input price moved from $0.24/M to $0.27/M (13% up). (details)
- DeepSeek V3 0324 output price moved from $0.90/M to $1.12/M (24% up). (details)
- Qwen3.5-4B (thinking) input price moved from $0.030/M to $0.33/M (1000% up). (details)
- Qwen3.5-4B (thinking) output price moved from $0.15/M to $1.00/M (570% up). (details)
- Qwen3.5-4B (no reasoning) input price moved from $0.030/M to $0.33/M (1000% up). (details)
- Qwen3.5-4B (no reasoning) output price moved from $0.15/M to $1.00/M (570% up). (details)
- R1 Distill Llama 70B output price moved from $1.10/M to $1.40/M (27% up). (details)
- Llama 3.1 70B input price moved from $0.56/M to $0.81/M (45% up). (details)
- Llama 3.1 70B output price moved from $0.56/M to $0.81/M (45% up). (details)
- Mistral Small 3.1 input price moved from $0.080/M to $0.11/M (38% up). (details)
- Mistral Small 3.1 output price moved from $0.14/M to $0.17/M (21% up). (details)
- Llama 3.2 Instruct 11B (Vision) input price moved from $0.35/M to $0.00/M (100% down). (details)
- Llama 3.2 Instruct 11B (Vision) output price moved from $0.35/M to $0.00/M (100% down). (details)
- …and 31 more.
New benchmark results
- MLCR: 9 new results, including Gemini 4 Neon, Gemini 4 Carbon, Gemini 4 Neon… (board)
- GameBench 2 by reasoning setting: 7 new results, including Gemini 4 Carbon, Gemini 4 Carbon, Gemini 4 Hydrogen… (board)
- Arena Text: 7 new results, including Gemini 4 Neon, DeepSeek V4 Pro 0424, DeepSeek V4 Flash 0420… (board)
- Arena Hard Prompts: 7 new results, including Gemini 4 Neon, DeepSeek V4 Pro 0424, DeepSeek V4 Flash 0420… (board)
- Arena Coding: 7 new results, including Gemini 4 Neon, DeepSeek V4 Pro 0424, DeepSeek V4 Flash 0420… (board)
- Arena Creative Writing: 7 new results, including Gemini 4 Neon, DeepSeek V4 Pro 0424, DeepSeek V4 Flash 0420… (board)
- Arena Instruction Following: 7 new results, including Gemini 4 Neon, DeepSeek V4 Pro 0424, DeepSeek V4 Flash 0420… (board)
- Arena Multi-turn: 7 new results, including Gemini 4 Neon, DeepSeek V4 Pro 0424, DeepSeek V4 Flash 0420… (board)
- Arena Longer Queries: 7 new results, including Gemini 4 Neon, DeepSeek V4 Pro 0424, DeepSeek V4 Flash 0420… (board)
- Arena Expert: 7 new results, including Gemini 4 Neon, DeepSeek V4 Pro 0424, DeepSeek V4 Flash 0420… (board)
- Arena Maths: 7 new results, including Kimi K3.1 Pro, DeepSeek V4 Pro 0424, DeepSeek V4 Flash 0420… (board)
- Epoch Capabilities Index: 6 new results, including GPT-4 Turbo Preview 1106, Gemini 1.5 Pro 001, Gemini 1.5 Pro 002… (board)
Score revisions
- Claude Fable 5.2 (max) on Arena Agent: 28.6 → 27.4. (details)
- Claude Fable 5.2 (max) on Arena Agent Task Completion: 19.4 → 22.5. (details)
- Gemini 4 Carbon (high) on Arena Agent Task Completion: 17.9 → 17.2. (details)
- Gemini 4 Carbon (high) on LiveBench Long Agentic: 72.7% → 71.8%. (details)
- GPT-6 Astra (max) on Arena Agent: 13.9 → 14.6. (details)
- GPT-6 Astra (max) on Arena Agent Task Completion: 13.6 → 14.9. (details)
- GPT-6 Astra (max) on LiveBench Long Agentic: 75.6% → 74.7%. (details)
- Claude Fable 5.1 (max) on Arena Agent Task Completion: 14.7 → 14.2. (details)
- GPT-6.1 Sol (max) on Arena Agent: 11.8 → 11.3. (details)
- GPT-6.1 Sol (max) on LiveBench Long Agentic: 65.6% → 64.7%. (details)
- Claude Opus 5.5 (high) on Arena Agent Task Completion: 11.7 → 12.6. (details)
- Gemini 4 Argon (high) on Arena Agent Task Completion: 13.2 → 12.6. (details)
- Claude Opus 5 (max) on Arena Agent Task Completion: 9.4 → 10.1. (details)
- GPT-5.6 Sol (xhigh) on Arena Agent Task Completion: 1.7 → 2.5. (details)
- Gemini 4 Hydrogen (high) on Arena Agent: 14.9 → 13.5. (details)
- Gemini 4 Hydrogen (high) on Arena Agent Task Completion: 16.8 → 15.7. (details)
- Gemini 4 Hydrogen (high) on Arena Maths: 1495 → 1504. (details)
- Claude Sonnet 5.5 (max) on LiveBench Long Agentic: 63.4% → 62.5%. (details)
- GPT-6.1 Terra (max) on Arena Agent: 11.3 → 10.5. (details)
- GPT-6.1 Terra (max) on Arena Agent Task Completion: 9.1 → 8.5. (details)
- …and 139 more.
Speed changes
- Claude Fable 5 output speed changed from 37 tok/s to 54 tok/s. (details)
- GPT-5.5 time to first token changed from 1.41 s to 2.17 s. (details)
- Claude Fable 5.1 time to first answer changed from 66.39 s to 41.37 s. (details)
- Claude Opus 4.7 time to first token changed from 0.68 s to 1.02 s. (details)
- Gemini 4 Hydrogen output speed changed from 86 tok/s to 136 tok/s. (details)
- GLM 5.3 time to first token changed from 0.24 s to 0.36 s. (details)
- DeepSeek V4 Pro 0424 time to first answer changed from 37.62 s to 54.66 s. (details)
- DeepSeek V4 Pro 0424 thinking time changed from 35.64 s to 52.76 s. (details)
- Inkling 2 Small time to first token changed from 0.32 s to 0.54 s. (details)
- Grok 4.7 time to first token changed from 1.76 s to 2.55 s. (details)
- MiniMax M2.7 thinking time changed from 9.15 s to 13.46 s. (details)
- MiMo-V2.5 time to first token changed from 3.00 s to 1.76 s. (details)
- MiMo-V2.5-Pro output speed changed from 21 tok/s to 52 tok/s. (details)
- MiMo-V2.5-Pro time to first token changed from 1.25 s to 1.77 s. (details)
- Nemotron 3 Ultra 550B A55B output speed changed from 61 tok/s to 93 tok/s. (details)
- …and 73 more.
Today's top five
- Claude Fable 5.2 (max) — 70.2
- Gemini 4 Carbon (high) — 68.6
- GPT-6.2 Astra (max) — 68.3
- Gemini 4 Neon (high) — 68.0
- GPT-6 Astra — 68.0
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.