MMODELYST
The score ledger

Changes

Every benchmark score on Modelyst is tracked in an append-only ledger. When a number moves between refreshes — a re-evaluation, a silent endpoint update, a harness change upstream — it lands here, with the old value, the new value, and the date. Last refresh: Jul 27, 2026.

Capability · price · speed drift — last 30 days
GPT-5 minicap -5.1-102 tok/sQwen3.5 4Bcap -5.0+6 tok/sQwen3.5 2Bcap -4.2price −0.04-36 tok/sGPT-5.5 Instant (May 2026)cap +2.7LFM2 2.6Bcap +2.4-336 tok/sQwen3.5 0.8Bcap -1.8price −0.02-30 tok/sMiMo-V2-Procap -1.8price −1.5-48 tok/sMercury 2cap -1.8+113 tok/sHy3-previewcap -1.8-41 tok/sDiffusionGemma 26B A4Bcap -1.8Mistral Medium 3.5cap -1.8-32 tok/sQwen3.5 Omni Pluscap -1.8-3 tok/s

Jul 27, 2026

35 changes · 34 first observations
Qwen3.5 2BGPQA Diamond45.6 → 32.5▼ 13.1
Qwen3.5 4BGPQA Diamond77.1 → 68▼ 9.1
Ministral 3 3BGPQA Diamond35.8 → 30.1▼ 5.7
Gemma 4 E4BGPQA Diamond57.6 → 52.2▼ 5.4
LFM2.5-8B-A1BGPQA Diamond51.3 → 46.6▼ 4.7
Gemma 4 E4BIFBench44.2 → 40.6▼ 3.6
Granite 4.0 350MGPQA Diamond20.3 → 23.7▲ 3.4
Granite 4.0 H 350MGPQA Diamond25.7 → 29▲ 3.3
Granite 4.0 MicroGPQA Diamond33.6 → 30.3▼ 3.3
Llama 3.2 Instruct 1BGPQA Diamond17.7 → 20.8▲ 3.1
Llama 3.2 Instruct 3BGPQA Diamond25.5 → 22.4▼ 3.1
LFM2.5-1.2B-InstructGPQA Diamond32.6 → 29.5▼ 3.1
Granite 4.0 350MMATH11.2 → 14▲ 2.8
LFM2.5-1.2B-InstructIFBench43.8 → 41▼ 2.8
Gemma 4 E2BGPQA Diamond43.3 → 40.5▼ 2.8
Granite 4.0 MicroIFBench24.8 → 22.1▼ 2.7
Ministral 3 3BIFBench26.8 → 24.1▼ 2.7
LFM2.5-8B-A1BIFBench55.6 → 53.3▼ 2.3
LFM2 2.6BGPQA Diamond29.2 → 31.5▲ 2.3
Gemma 4 E2BIFBench38 → 36▼ 2.0
Llama 3.2 Instruct 3BMATH48.9 → 47▼ 1.9
Qwen3.5 0.8BMATH11.1 → 9.3▼ 1.8
Qwen3.5 4BIFBench52 → 50.2▼ 1.8
Granite 4.0 350MIFBench16.8 → 15.1▼ 1.7
Granite 4.0 H 1BGPQA Diamond26.3 → 24.6▼ 1.7
LFM2 2.6BMATH76.8 → 75.3▼ 1.5
Qwen3.5 2BIFBench31.5 → 30.4▼ 1.1
Llama 3.2 Instruct 1BMATH26.6 → 27.7▲ 1.1
Granite 4.0 H 1BIFBench26.2 → 25.1▼ 1.1
Llama 3.2 Instruct 3BIFBench26.2 → 25.2▼ 1.0
Llama 3.2 Instruct 1BIFBench22.7 → 23.5▲ 0.8
LFM2 2.6BIFBench26.5 → 25.7▼ 0.8
Qwen3.5 0.8BIFBench21.6 → 20.9▼ 0.7
Qwen3.5 0.8BGPQA Diamond11.4 → 12▲ 0.6
Granite 4.0 H 350MIFBench17.6 → 17.1▼ 0.5

Jul 20, 2026

10 changes · 19 first observations
Llama 3.2 Instruct 1BMATH14 → 26.6▲ 12.6
LFM2 2.6BIFBench19.5 → 26.5▲ 7.0
Granite 4.0 350MGPQA Diamond26.1 → 20.3▼ 5.8
Llama 3.2 Instruct 1BGPQA Diamond19.6 → 17.7▼ 1.9
LFM2 2.6BGPQA Diamond30.6 → 29.2▼ 1.4
Granite 4.0 350MIFBench15.9 → 16.8▲ 0.9
Qwen3.5 0.8BGPQA Diamond11.1 → 11.4▲ 0.3
Llama 3.2 Instruct 1BIFBench22.8 → 22.7▼ 0.1
Qwen3.5 0.8BIFBench21.5 → 21.6▲ 0.1
North Mini CodeHumanity's Last Exam9.9 → 10▲ 0.1

Jul 6, 2026

14 changes · 16 first observations
GPT-5 miniLiveCodeBench83.8 → 69.2▼ 14.6
GPT-5.5 Instant (May 2026)AA Long-Context Reasoning55.7 → 64▲ 8.3
GPT-5 miniAIME90.7 → 85▼ 5.7
GPT-5 miniHumanity's Last Exam19.7 → 14.6▼ 5.1
GPT-5 miniTerminal-Bench Hard33.3 → 28.8▼ 4.5
GPT-5 miniIFBench75.4 → 71.2▼ 4.2
GPT-5 miniτ²-bench68.4 → 71.1▲ 2.7
GPT-5 miniGPQA Diamond82.8 → 80.3▼ 2.5
GPT-5.5 Instant (May 2026)GPQA Diamond84.6 → 82.3▼ 2.3
GPT-5 miniAA Long-Context Reasoning68 → 66▼ 2.0
GPT-5 miniSciCode39.2 → 41▲ 1.8
GPT-5.5 Instant (May 2026)Humanity's Last Exam20.3 → 18.6▼ 1.7
GPT-5.5 Instant (May 2026)SciCode50.3 → 48.6▼ 1.7
GPT-5 miniMMLU-Pro83.7 → 82.8▼ 0.9

Jun 29, 2026

10 changes · 10 first observations
Nova 2.0 LiteAIME94.3 → 88.7▼ 5.6
Nova 2.0 LiteLiveCodeBench71.1 → 66.3▼ 4.8
Nova 2.0 LiteGPQA Diamond81.1 → 76.8▼ 4.3
Nova 2.0 LiteAA Long-Context Reasoning55.3 → 58.3▲ 3.0
Nova 2.0 Liteτ²-bench72.8 → 75.7▲ 2.9
Nova 2.0 LiteHumanity's Last Exam10.9 → 8.6▼ 2.3
Nova 2.0 LiteIFBench70.7 → 68.5▼ 2.2
Nova 2.0 LiteTerminal-Bench Hard16.7 → 17.4▲ 0.7
Nova 2.0 LiteMMLU-Pro81.8 → 81.3▼ 0.5
Nova 2.0 LiteSciCode36.9 → 36.8▼ 0.1

Source of each value: see the score's provenance on its model page. Scores via Artificial Analysis are medians across providers; changes can reflect re-evaluation, endpoint updates, or methodology changes upstream.