Coding

SciCode leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

As of 2026-10-07, the highest measured score on SciCode is 66.9% by Claude Opus 5.5 (max with fallback).

#	Model	Score	Source
1	Claude Opus 5.5 (max with fallback)	66.9%	measured
2	Claude Opus 5.5 (xhigh with fallback)	65.0%	measured
3	Claude Fable 5.1 (max with fallback)	63.1%	measured
4	Gemini 4 Argon (high)	61.8%	measured
5	Claude Sonnet 5.5 (max with fallback)	61.0%	measured
6	Claude Fable 5 (with fallback)	61.0%	measured
7	Claude Fable 5.1 (xhigh with fallback)	60.9%	measured
8	MiMo-V2.6-Pro	60.9%	measured
9	Claude Opus 5.5 (high with fallback)	60.4%	measured
10	Gemini 3.7 Flash (medium)	59.8%	measured
11	Kimi K3 (max)	59.5%	measured
12	GLM-5.3 (max)	59.0%	measured
13	Step 5 Preview	58.9%	measured
14	Muse Spark 1.3 (max)	58.8%	measured
15	GPT-6 Sol (max)	57.6%	measured
16	Grok 4.7 (xhigh)	57.4%	measured
17	Gemini 3.8 Flash (high)	56.6%	measured
18	GPT-6 Astra (max)	56.5%	measured
19	GPT-6 Luna (max)	54.6%	measured
20	GPT-6.1 Sol (max)	54.2%	measured
21	Mistral Large 4 Preview	54.2%	measured
22	Qwen3.8 Max (0902)	52.1%	measured
23	DeepSeek V4.1 Flash (max)	51.9%	measured
24	GLM-5.3-Flash	51.6%	measured
25	MiniMax-M3	47.1%	measured
26	Inkling (xhigh)	47.0%	measured
27	Qwen3.8 27B (xhigh)	46.6%	measured
28	Muse Glimmer (high)	44.9%	measured
29	K2 Horizon 375B A23B	42.9%	measured
30	Nemotron 3 Ultra	40.3%	measured

All leaderboards