BenchLeader

Claude Opus 5.5, GPT-6 Sol + 10 new AI models: 21 Sep–27 Sep

12 new models incl. Claude Opus 5.5, GPT-6 Sol and Grok 4.7, MiMo-V2.6-Pro up 3.5, 23 price cuts, 2211 new benchmark results: what moved on the BenchLeader Index this week.

The week of 21 Sep 2026 to 27 Sep 2026, from 7 daily updates.

Three things that happened

  • Biggest riser: MiMo-V2.6-Pro, up 3.5 points to 61.3 on the BenchLeader Index as new results landed.
  • 12 new models arrived: GLM 5.3 Prime, Qwen3.8 Max Prime, GPT-6 Sol, GPT-6 Luna, gpt-6-luna-pro and more.
  • Busiest benchmark: AA-Omniscience v1.1: accuracy with 498 new results; 2211 new results were published in all.

Index movers

Change in BenchLeader Index, 2026-09-21 to 2026-09-27; best configuration per model.

| Model | Provider | 21 Sep 2026 | 27 Sep 2026 | Change | |---|---|---:|---:|---:| | Mistral Small 3.1 | Mistral AI | 39.6 | 36.0 | -3.6 | | MiMo-V2.6-Pro | Xiaomi | 57.8 | 61.3 | +3.5 | | Muse Spark 1.1 | Meta | 65.2 | 62.3 | -2.9 | | GPT-6 Luna | OpenAI | 56.8 | 59.6 | +2.8 | | DeepSeek V4.1 Flash | DeepSeek | 58.9 | 61.3 | +2.4 | | GPT-6 Sol | OpenAI | 62.0 | 64.1 | +2.1 | | Claude Haiku 4.5 | Anthropic | 48.2 | 50.3 | +2.1 | | GPT-5.2 | OpenAI | 61.3 | 63.3 | +2.0 | | DeepSeek V4 Pro | DeepSeek | 58.3 | 56.4 | -1.9 | | Command R+ | Cohere | 35.2 | 33.3 | -1.9 |

New models

  • GLM 5.3 Prime (Zhipu AI) is now listed at $4.30/M blended; no independent results yet. (details)
  • Qwen3.8 Max Prime (Alibaba) is now listed at $6.00/M blended; no independent results yet.
  • GPT-6 Sol (no reasoning) (OpenAI) appeared with 5 benchmark results, entering the BenchLeader Index at #249. (details)
  • GPT-6 Luna (no reasoning) (OpenAI) appeared with 8 benchmark results, entering the BenchLeader Index at #465. (details)
  • GPT-6 Luna Pro (OpenAI) is now listed at $0.20/M blended; no independent results yet. (details)
  • GPT-6 Sol Pro (OpenAI) is now listed at $4.00/M blended; no independent results yet. (details)
  • Claude Opus 5.5 (low) (Anthropic) appeared with 5 benchmark results, entering the BenchLeader Index at #39. (details)
  • MiMo-V2.6-Pro (Xiaomi) appeared with 4 benchmark results. (details)
  • GPT-6 Astra Pro Max (OpenAI) appeared with 1 benchmark result.
  • SWE-2 (max) (Cognition) appeared with 1 benchmark result.
  • Grok 4.7 (xhigh) (SpaceXAI) appeared with 6 benchmark results, entering the BenchLeader Index at #72. (details)
  • Dots3 Note (max) (Dots Studio) appeared with 1 benchmark result.

Price changes

  • Llama 3.1 Nemotron 70B Instruct input price moved from $1.20/M to $0.00/M (100% down).
  • Nemotron Nano 12B v2 VL (thinking) input price moved from $0.20/M to $0.00/M (100% down).
  • Devstral Small 2 input price moved from $0.10/M to $0.00/M (100% down).
  • Ministral 3 14B cost per task moved from 0.1 to 0 (86% down).
  • Grok 4.1 (thinking) input price moved from $1.25/M to $0.20/M (84% down).
  • Grok 4 Fast (thinking) input price moved from $1.25/M to $0.20/M (84% down).
  • Qwen3.5 Flash input price moved from $0.17/M to $0.03/M (83% down).
  • Ministral 3 8B cost per task moved from 0.1 to 0 (83% down).
  • GLM 5V Turbo output price moved from $22.00/M to $4.00/M (82% down).
  • Qwen3 Max (max) output price moved from $7.75/M to $1.43/M (81% down).
  • Ministral 3 3B cost per task moved from 0 to 0 (81% down).
  • Grok 3 mini (thinking) output price moved from $2.50/M to $0.50/M (80% down).
  • Mistral Small 3 output price moved from $0.30/M to $0.08/M (73% down).
  • Mistral Small 4 (thinking) cost per task moved from 0 to 0 (67% down).
  • Qwen3.5 Plus 2026-04-20 (thinking) output price moved from $1.72/M to $0.69/M (60% down).

Top five as of 8 Nov 2026

  1. Claude Opus 5.5 — 69.4
  2. Gemini 4 Carbon — 69.3
  3. GPT-6 Astra — 68.3
  4. Claude Fable 5.1 — 67.3
  5. GPT-6.1 Sol — 67.3

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

Generated from the week's daily updates; every line links to the page where the numbers and their sources can be checked.

Head to head

Get the weekly review by email

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