BenchLeader
Menu

State of the leaderboard: week 39, 21–27 September 2026

Complete week: data as of , from 7 daily updates

The week in 6 sentences

  • In week 39 of 2026 (21–27 September 2026), GPT-6 Astra (OpenAI) held the #1 spot on the BenchLeader Index at 69.4, 0.6 points ahead of Claude Fable 5.1 (68.8).
  • 2 models entered the top 10 (Claude Opus 5.5 at #6 and GPT-5.4 at #10), pushing out Muse Spark 1.1 and Kimi K3; 5 other models changed places.
  • 12 new models arrived; the highest-ranked is Claude Opus 5.5 (Anthropic) at #6 with an index of 66.0.
  • The biggest riser was MiMo-V2.6-Pro, up 3.5 points to 61.3; the biggest faller Mistral Small 3.1, down 3.6 to 36.0.
  • 16 models cut a list price and 45 models raised one; the largest cut was labs-devstral-small's input price, 100% lower.
  • 2,211 new benchmark results were published, the most on AA-Omniscience v1.1: accuracy (498).

Top 10 at the end of the week

The BenchLeader Index, one row per model at its best reasoning effort, against the table at the end of the previous week. Left the top 10: Muse Spark 1.1, Kimi K3.

#ModelProviderIndexLast weekMove
1GPT-6 AstraOpenAI69.41–
2Claude Fable 5.1Anthropic68.82–
3Claude Opus 5Anthropic67.64↑1
4Claude Fable 5Anthropic67.13↓1
5GPT-5.6 SolOpenAI66.55–
6Claude Opus 5.5Anthropic66.0–new
7GPT-5.5OpenAI65.86↓1
8Muse Spark 1.3Meta65.77↓1
9Muse SparkMeta65.68↓1
10GPT-5.4OpenAI65.3–new

New models and their first results

  • Grok 4.7 (SpaceXAI): #48 on the index at 59.9. First results: 25.8% on Terminal-Bench 4.0 (AA); 47.5% on AA-Omniscience v1.1: accuracy; 70.7% on AA-Omniscience v1.1: non-hallucination; 65.6% on AutomationBench.
  • Dots3 Note (Dots Studio): not enough independent results for an index yet
  • Claude Opus 5.5 (Anthropic): #6 on the index at 66.0. First results: 66.7% on MLCR; 98.5% on ARC-AGI-1; 93.3% on ARC-AGI-2; 82.1% on LiveBench 1.0.
  • MiMo-V2.6-Pro (Xiaomi): #36 on the index at 61.3. First results: 18.3% on MLCR; 67.8% on Terminal-Bench 2.1 (Vals); 59.7 on Vals Index; 43.0% on Code Migration.
  • gpt-6-astra-promax: not enough independent results for an index yet
  • SWE-2 (Cognition): not enough independent results for an index yet
  • GPT-6 Sol (OpenAI): #15 on the index at 64.1. First results: 53.3% on EBR-bench; 1686 on Arena WebDev; 49.4% on ITBench SRE (AA); 1458 on Arena Text.
  • GPT-6 Luna (OpenAI): #54 on the index at 59.6. First results: 1593 on Arena WebDev; 1442 on Arena Text; 1468 on Arena Hard Prompts; 1515 on Arena Coding.
  • gpt-6-luna-pro: not enough independent results for an index yet
  • gpt-6-sol-pro: not enough independent results for an index yet
  • GLM 5.3 Prime (Zhipu AI): not enough independent results for an index yet
  • Qwen3.8 Max Prime (Alibaba): not enough independent results for an index yet

Biggest movers on the index

ModelProvider21 Sep 202627 Sep 2026Change
Mistral Small 3.1Mistral AI39.636.0-3.6
MiMo-V2.6-ProXiaomi57.861.3+3.5
Muse Spark 1.1Meta65.262.3-2.9
GPT-6 LunaOpenAI56.859.6+2.8
DeepSeek V4.1 FlashDeepSeek58.961.3+2.4
GPT-6 SolOpenAI62.064.1+2.1
Claude Haiku 4.5Anthropic48.250.3+2.1
GPT-5.2OpenAI61.363.3+2.0
DeepSeek V4 ProDeepSeek58.356.4-1.9
Command R+Cohere35.233.3-1.9

Price changes

Cheaper (16)

  • Devstral Small 2 input price moved from $0.10/M to $0.00/M (100% down).
  • 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).
  • 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 $0.029/M (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).
  • 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 $0.080/M (73% down).
  • Qwen3.5 Plus 2026-04-20 (thinking) output price moved from $1.72/M to $0.69/M (60% down).
  • Gemini Flash Latest output price moved from $9.00/M to $3.75/M (58% down).
  • …and 4 more.

More expensive (45)

  • Qwen-VL OCR input price moved from $0.043/M to $0.72/M (1574% up).
  • Qwen-MT Plus output price moved from $0.78/M to $7.37/M (851% up).
  • Grok 3 output price moved from $2.50/M to $20.00/M (700% up).
  • Qwen3-Omni Flash input price moved from $0.058/M to $0.43/M (641% up).
  • Grok 4 output price moved from $2.50/M to $15.00/M (500% up).
  • Qwen Vl (max) output price moved from $0.57/M to $3.20/M (457% up).
  • Qwen2.5 7B Instruct output price moved from $0.14/M to $0.70/M (386% up).
  • Qwen3 Max (max) output price moved from $1.43/M to $6.00/M (318% up).
  • Qwen-Plus output price moved from $0.29/M to $1.20/M (318% up).
  • QwQ Plus output price moved from $0.57/M to $2.40/M (318% up).
  • Mistral Large 2 input price moved from $0.50/M to $2.00/M (300% up).
  • Qwen3.5 Plus 2026-04-20 (thinking) output price moved from $0.69/M to $2.40/M (249% up).
  • …and 33 more.

List prices per million tokens; each model's full price history is on its page and the LLM API pricing tracker.

Other weeks

← Week 40, 28 Sep–4 Oct 2026 · Week 38, 14–20 Sep 2026 → · The weekly review on the blog

Benchmark categories: AI agent benchmarks · AI coding benchmarks · Reasoning and math benchmarks · Multimodal AI benchmarks · New AI model releases

How this is built: the BenchLeader Index is recomputed with every update from independent benchmark publishers (see the methodology), and a maker's own launch numbers are never used. Past ranks are rebuilt from the daily index history of the models listed today.

Cite as: BenchLeader, “AI leaderboard week 39, 21–27 Sep 2026: who was #1”, data as of 27 Sep 2026, https://www.benchleader.com/state/2026-w39.