Benchmark

SciCode

SciCode is a research coding benchmark curated by scientists that challenges language models to code solutions for scientific problems. It contains 338…

Modality
text
Categories
math, physics, reasoning, biology, chemistry, code
Openness
unknown
Source
llm_stats
Reported scores
21

Reported scores

llm_stats

ModelOrganizationReported valueReportedEvidence
Command A+Cohere0.382026-05-20evidence
Gemini 3.1 ProGoogle0.592026-02-19evidence
GLM-4.5Z.ai0.4172025-07-28evidence
GLM-4.5-AirZ.ai0.3732025-07-28evidence
Inkling-SmallThinking Machines Lab0.4872026-07-30evidence
Kimi K2-Thinking-0905Moonshot AI0.4482025-09-05evidence
Kimi K2.5Moonshot AI0.4872026-01-27evidence
Kimi K2.6Moonshot AI0.5222026-04-20evidence
Mercury 2Inception0.382026-02-24evidence
MiniMax M2MiniMax0.362025-10-27evidence
MiniMax M2.1MiniMax0.392025-12-23evidence
Muse Glimmer-30BMeta0.4362026-08-10evidence
Nemotron 3 Nano (30B A3B)NVIDIA0.3332025-12-15evidence
Nemotron 3 Super (120B A12B)NVIDIA0.42052026-03-11evidence
Nemotron 3 Ultra (550B A55B)NVIDIA0.4462026-06-04evidence
Nemotron 3.5 Lightning (30B A3B)NVIDIA0.3262026-08-11evidence
North Mini Code 1.0Cohere0.3822026-06-09evidence
Qwen3.7 MaxQwen0.5352026-05-19evidence
Qwen3.7-PlusQwen0.5132026-05-31evidence
Seed 2.1 ProByteDance0.5982026-06-24evidence
Seed 2.1 TurboByteDance0.5782026-06-24evidence

Scores are partitioned by the source that reported them and are never merged into a single cross-source ranking, because the sources measure different things and say so.