Daily digest, 24 Sept 2026

BenchLeader refreshed 1550 models from 22 of 22 sources. 2 new models, 12 price changes, 21 new or updated benchmark results across 22 sources.

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. (details)

Price changes

• Trinity Large (thinking) output price moved from $0.90/M to $0.80/M (11% down). (details)
• Llama 3.1 Nemotron 70B Instruct input price moved from $1.20/M to $0.00/M (100% down). (details)
• Llama 3.1 Nemotron 70B Instruct output price moved from $1.20/M to $0.00/M (100% down). (details)
• Nemotron Nano 12B v2 VL (thinking) input price moved from $0.20/M to $0.00/M (100% down). (details)
• Nemotron Nano 12B v2 VL (thinking) output price moved from $0.60/M to $0.00/M (100% down). (details)
• Mistral Small 3.1 input price moved from $0.10/M to $0.11/M (10% up). (details)
• Mistral Small 3.1 output price moved from $0.30/M to $0.17/M (43% down). (details)
• Mistral Small 3.1 cost per task moved from 0 to 0 (7% up). (details)
• Mistral Small 3 input price moved from $0.10/M to $0.05/M (50% down). (details)
• Mistral Small 3 output price moved from $0.30/M to $0.08/M (73% down). (details)
• Mistral 7B v0.3 input price moved from $0.25/M to $0.15/M (40% down). (details)
• Mistral 7B v0.3 output price moved from $0.25/M to $0.20/M (20% down). (details)

New benchmark results

• LMArena WebDev: 2 new results, including GPT-6 Sol, Claude Opus 5.5 (board)
• CyberBench: 2 new results, including MiMo-V2.6-Pro, MiMo-V2.6-Flash (board)
• MysteryMechanism: 2 new results, including MiMo-V2.6-Pro, MiMo-V2.6-Flash (board)
• Public Benefits Bench: 2 new results, including MiMo-V2.6-Pro, MiMo-V2.6-Flash (board)
• Tax Agent Bench: 2 new results, including MiMo-V2.6-Pro, MiMo-V2.6-Flash (board)
• Terminal-Bench 4.0 (Vals): 2 new results, including MiMo-V2.6-Pro, MiMo-V2.6-Flash (board)
• EBR-bench: 1 new result, including GPT-6 Sol (board)
• IOI: 1 new result, including MiMo-V2.6-Pro (board)
• ProgramBench: 1 new result, including MiMo-V2.6-Pro (board)
• Terminal-Bench 4.0 (AA): 1 new result, including Apodex 1.1 (board)
• Terminal-Bench Science: 1 new result, including MiMo-V2.6-Flash (board)

Score revisions

• Kimi K3 (max) on AA-Briefcase: 1510 → 1504. (details)
• GLM 5.3 (max) on AA-Briefcase: 1525 → 1516. (details)
• Grok 4.6 (high) on LMArena WebDev: 1616 → 1624. (details)
• MiMo-V2.6-Pro on Finance Agent v2: 58.3% → 57.3%. (details)

Speed changes

• GPT-6 Sol time to first token changed from 2.35 s to 4.65 s. (details)
• GPT-5.1-Codex output speed changed from 24 tok/s to 35 tok/s. (details)
• Ling 3.0 Flash VL time to first token changed from 0.90 s to 1.41 s. (details)
• Kimi K2 Thinking time to first token changed from 1.84 s to 0.99 s. (details)
• Nemotron 3 Nano 30B A3B output speed changed from 129 tok/s to 184 tok/s. (details)
• Mistral Medium 3 output speed changed from 153 tok/s to 65 tok/s. (details)
• Mistral Small 3.2 output speed changed from 157 tok/s to 24 tok/s. (details)
• Nova Pro output speed changed from 30 tok/s to 12 tok/s. (details)
• Qwen3.5-9B output speed changed from 26 tok/s to 37 tok/s. (details)
• Nemotron Nano 12B v2 VL time to first answer changed from 2.92 s to 1.10 s. (details)
• Nemotron Nano 12B v2 VL time to first token changed from 2.92 s to 1.10 s. (details)
• Nemotron Nano 12B v2 VL response time changed from 5.60 s to 3.56 s. (details)
• Mistral Small 4 output speed changed from 154 tok/s to 93 tok/s. (details)
• Hy-MT2-1.8B output speed changed from 98 tok/s to 38 tok/s. (details)
• Hy-MT2-1.8B time to first token changed from 0.34 s to 0.59 s. (details)
• …and 5 more.

Today's top five

1. GPT-6 Astra — 71.4
2. GPT-6 Astra — 71.1
3. Claude Fable 5.1 — 70.9
4. Claude Fable 5.1 — 70.8
5. Claude Fable 5.1 — 70.8

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.