mimo-v2-omni-0327 added, Qwen3 added (10 Sept 2026)

BenchLeader refreshed 1255 models from 19 of 19 data feeds (benchmark publishers, aggregators and price lists). 78 new models, 2 top-10 rank changes, 112 price changes, 223 new or updated benchmark results across 19 data feeds.

New models

MiMo-V2-Omni-0327 (Xiaomi) appeared with 6 benchmark results, entering the BenchLeader Index at #135. (details)
Qwen3 (max) (Alibaba) appeared with 5 benchmark results, entering the BenchLeader Index at #192. (details)
DeepSeek V4 Flash 0731 (DeepSeek) appeared with 5 benchmark results, entering the BenchLeader Index at #277. (details)
Kimi K2 0905 (Moonshot AI) appeared with 11 benchmark results, entering the BenchLeader Index at #280. (details)
Qwen3 235B A22B 2507 (Alibaba) appeared with 10 benchmark results, entering the BenchLeader Index at #281. (details)
Gemini 2.5 Flash 09 2025 (Google) appeared with 20 benchmark results, entering the BenchLeader Index at #283. (details)
DeepSeek R1 0528 (DeepSeek) appeared with 13 benchmark results, entering the BenchLeader Index at #313. (details)
Qwen3 30B A3B 2507 (Alibaba) appeared with 11 benchmark results, entering the BenchLeader Index at #377. (details)
GPT 4 0125 (OpenAI) appeared with 5 benchmark results, entering the BenchLeader Index at #396.
Magistral Medium 1.2 (Mistral AI) appeared with 16 benchmark results, entering the BenchLeader Index at #403. (details)
GPT 4 0314 (OpenAI) appeared with 5 benchmark results, entering the BenchLeader Index at #485. (details)
qwen3-4b-instruct-2507 (Alibaba) appeared with 8 benchmark results, entering the BenchLeader Index at #501. (details)
Magistral Small 1.2 (Mistral AI) appeared with 17 benchmark results, entering the BenchLeader Index at #505. (details)
Mistral Large (Mistral AI) appeared with 6 benchmark results, entering the BenchLeader Index at #514. (details)
GPT 3.5 Turbo 1106 (OpenAI) appeared with 5 benchmark results, entering the BenchLeader Index at #545. (details)
GPT 4 0613 (OpenAI) appeared with 7 benchmark results, entering the BenchLeader Index at #549. (details)
GPT 3.5 Turbo 0125 (OpenAI) appeared with 8 benchmark results, entering the BenchLeader Index at #607. (details)
Mistral Small 3.1 (Mistral AI) appeared with 12 benchmark results, entering the BenchLeader Index at #616. (details)
Mistral Small 2402 (Mistral AI) appeared with 7 benchmark results, entering the BenchLeader Index at #623. (details)
DeepSeek V4.1 Flash (DeepSeek) appeared with 4 benchmark results. (details)
Ling-3.0-flash-VL (Ant Group) appeared with 4 benchmark results. (details)
Mistral Small 3.1 24B 2503 (Mistral AI) appeared with 4 benchmark results. (details)
Athene 70B 0725 (Nexusflow) appeared with 3 benchmark results. (details)
Command R (08-2024) (Cohere) appeared with 3 benchmark results.
Command R+ (08-2024) (Cohere) appeared with 3 benchmark results.
Deepseek v2 5 1210 (DeepSeek) appeared with 3 benchmark results. (details)
Devstral 2 2512 (Mistral AI) appeared with 3 benchmark results. (details)
Ernie 5.0 0110 (Baidu) appeared with 3 benchmark results. (details)
Ernie 5.0 1022 (Baidu) appeared with 3 benchmark results. (details)
Ernie 5.0 1203 (Baidu) appeared with 3 benchmark results. (details)
…and 48 more.

Movement in the top 10

Claude Fable 5.1 (max) moved from #4 to #3 in the BenchLeader Index. (details)
Claude Fable 5.1 (xhigh) moved from #3 to #4 in the BenchLeader Index. (details)

Price changes

GPT-6 Astra (max) input price moved from $11.00/M to $10.00/M (9% down). (details)
GPT-6 Astra (max) output price moved from $55.00/M to $50.00/M (9% down). (details)
GPT-5.6 Sol (max) input price moved from $4.40/M to $2.00/M (55% down). (details)
GPT-5.6 Sol (max) output price moved from $22.00/M to $10.00/M (55% down). (details)
GPT-6 Astra (no reasoning) input price moved from $11.00/M to $10.00/M (9% down). (details)
GPT-6 Astra (no reasoning) output price moved from $55.00/M to $50.00/M (9% down). (details)
Grok 4.6 (medium) cached input price moved from $0.55/M to $0.50/M (9% down). (details)
GPT-5.5 (xhigh) input price moved from $5.50/M to $5.00/M (9% down). (details)
GPT-5.5 (xhigh) output price moved from $33.00/M to $30.00/M (9% down). (details)
Kimi K3 (max) input price moved from $0.00/M to $3.00/M (300000000000% up). (details)
Kimi K3 (max) output price moved from $0.00/M to $15.00/M (1500000000000% up). (details)
Grok 4.6 (xhigh) cached input price moved from $0.55/M to $0.50/M (9% down). (details)
GPT-5.4 (xhigh) input price moved from $2.75/M to $2.50/M (9% down). (details)
GPT-5.4 (xhigh) output price moved from $16.50/M to $15.00/M (9% down). (details)
Muse Spark 1.2 (xhigh) input price moved from $1.25/M to $0.10/M (92% down). (details)
Muse Spark 1.2 (xhigh) output price moved from $4.25/M to $0.20/M (95% down). (details)
Grok 4.6 (low) cached input price moved from $0.55/M to $0.50/M (9% down). (details)
DeepSeek V4 Flash input price moved from $0.14/M to $0.15/M (7% up). (details)
DeepSeek V4 Flash output price moved from $0.28/M to $0.60/M (114% up). (details)
DeepSeek V4 Flash cached input price moved from $0.0028/M to $0.0030/M (7% up). (details)
GPT-5.4 (high) input price moved from $2.75/M to $2.50/M (9% down). (details)
GPT-5.4 (high) output price moved from $16.50/M to $15.00/M (9% down). (details)
Grok 4.6 (high) input price moved from $2.20/M to $2.00/M (9% down). (details)
Grok 4.6 (high) output price moved from $6.60/M to $6.00/M (9% down). (details)
Grok 4.6 (high) cached input price moved from $0.55/M to $0.50/M (9% down). (details)
GPT-5.6 Terra (max) input price moved from $2.20/M to $2.00/M (9% down). (details)
GPT-5.6 Terra (max) output price moved from $13.20/M to $12.00/M (9% down). (details)
GLM 5.2 (max) input price moved from $1.10/M to $0.28/M (75% down). (details)
GLM 5.2 (max) output price moved from $3.85/M to $0.88/M (77% down). (details)
Deepseek v4 Flash Vision cached input price moved from $0.0028/M to $0.0030/M (7% up). (details)
…and 82 more.

New benchmark results

SciCode: 6 new results, including qwen3-8@max, DeepSeek V4 Flash, Mistral Large 3… (board)
Arena Text: 6 new results, including Qwen3.6 Max, Qwen3.7, DeepSeek V4 Flash… (board)
Arena Hard Prompts: 6 new results, including Qwen3.6 Max, Qwen3.7, DeepSeek V4 Flash… (board)
Arena Coding: 6 new results, including Qwen3.6 Max, Qwen3.7, DeepSeek V4 Flash… (board)
CorpFin: 4 new results, including Qwen3.6 Max, Mistral Large 3, Mistral Medium 3… (board)
SWE-bench (Vals): 3 new results, including Qwen3.6 Max, Mistral Large 3, Command A (board)
AIME (Vals): 3 new results, including Mistral Large 3, Mistral Medium 3, Command A (board)
LiveCodeBench: 3 new results, including Mistral Large 3, Mistral Medium 3, Command A (board)
MMLU-Pro: 3 new results, including Mistral Large 3, Mistral Medium 3, Command A (board)
LegalBench: 3 new results, including Mistral Large 3, Mistral Medium 3, Command A (board)
TaxEval: 3 new results, including Mistral Large 3, Mistral Medium 3, Command A (board)
MedQA: 3 new results, including Mistral Large 3, Mistral Medium 3, Command A (board)

Score revisions

GPT-6 Astra (max) on ARC-AGI-3: 62.7% → 98.6%. (details)
GPT-6 Astra (high) on ARC-AGI-3: 54.8% → 99.9%. (details)
GPT-6 Astra (xhigh) on ARC-AGI-3: 59.3% → 98.4%. (details)
GPT-6 Astra (medium) on ARC-AGI-3: 38.6% → 98.4%. (details)
GPT-6 Astra (low) on ARC-AGI-3: 17.5% → 98%. (details)
GPT-6 Astra (no reasoning) on ARC-AGI-3: 35.2% → 96.7%. (details)
Qwen3 8 (max) on LMArena WebDev: 1685 → 1670. (details)
DeepSeek V4 Pro on Epoch Capabilities Index: 155.4 → 149.1. (details)
DeepSeek V4 Pro on LiveBench: 77.4% → 71.6%. (details)
DeepSeek V4 Pro on LiveBench Reasoning: 85.8% → 82.7%. (details)
DeepSeek V4 Pro on LiveBench Coding: 77.2% → 70%. (details)
DeepSeek V4 Pro on LiveBench Agentic Coding: 55% → 42.6%. (details)
DeepSeek V4 Pro on LiveBench Mathematics: 95.1% → 90.7%. (details)
DeepSeek V4 Pro on LiveBench Data Analysis: 79.2% → 74.5%. (details)
DeepSeek V4 Pro on LiveBench Language: 82.1% → 78.1%. (details)
DeepSeek V4 Flash (high) on WeirdML: 57.1% → 43.8%. (details)
Qwen3.8-Flash-Next on AA Intelligence Index: 42.2 → 39.9. (details)
DeepSeek V4 Flash on SimpleBench: 61.1% → 46.3%. (details)
DeepSeek V4 Flash on Epoch Capabilities Index: 154.5 → 146.1. (details)
DeepSeek V4 Flash on LiveBench: 74.2% → 65.5%. (details)
…and 125 more.

Speed changes

GPT-6 Astra output speed changed from 54 tok/s to 27 tok/s. (details)
Claude Fable 5.1 output speed changed from 67 tok/s to 40 tok/s. (details)
Claude Fable 5 output speed changed from 63 tok/s to 37 tok/s. (details)
GPT-5.6 Sol output speed changed from 68 tok/s to 38 tok/s. (details)
GPT-5.5 output speed changed from 88 tok/s to 50 tok/s. (details)
Gemini 3.8 Flash output speed changed from 273 tok/s to 90 tok/s. (details)
Gemini 3 Flash time to first token changed from 0.85 s to 1.22 s. (details)
GPT-5.4 output speed changed from 130 tok/s to 34 tok/s. (details)
Muse Spark 1.2 output speed changed from 219 tok/s to 136 tok/s. (details)
GLM 5.3 Flash time to first token changed from 1.70 s to 2.43 s. (details)
Gemini 3.7 Flash output speed changed from 297 tok/s to 121 tok/s. (details)
Gemini 3.5 Flash output speed changed from 209 tok/s to 112 tok/s. (details)
DeepSeek V4 Pro time to first answer changed from 27.09 s to 70.73 s. (details)
DeepSeek V4 Flash output speed changed from 118 tok/s to 44 tok/s. (details)
DeepSeek V4 Flash time to first answer changed from 18.16 s to 11.74 s. (details)
…and 100 more.

Today's top five

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.