Benchmarks of Step316
BenchmarkByScoreRank
Knowledge15
Multimodal1
No benchmark matches
—per task?
self-host onlyprice
—output speed?
—context
Ranking?
Alternatives?
ModelScoreCostPer taskWhat changes
BetterGPT-6 Lunamaxwins 53%—$0.068Function calling only with none · Chat Completions+3
OpenGLM 5.3maxwins 65%—$2.01Memory (4-bit): 193 GB → 452 GB+5
BetterClaude Haiku 5.5highwins 55%—$0.079Remove sampling parameters+5BetterGPT-5.6 Lunaxhighwins 60%—$0.085BetterGPT-5.6 Terraxhighwins 61%—$0.63Prompts with >272K input tokensModelScoreCostPer taskWhat changes
Nothing here yet.
ModelScoreCostPer taskWhat changes
GPT-6 Lunamaxwins 53%—$0.068Function calling only with none · Chat Completions+3Claude Haiku 5.5highwins 55%—$0.079Remove sampling parameters+5GPT-5.6 Lunaxhighwins 60%—$0.085GPT-5.6 Terraxhighwins 61%—$0.63Prompts with >272K input tokensGPT-6.1 Solmaxwins 61%—$0.72Remove unsupported parameters+5ModelScoreCostPer taskWhat changes
GLM 5.3maxOpenwins 65%—$2.01Memory (4-bit): 193 GB → 452 GB+5GLM 5.1unstatedOpenwins 64%—$1.40 / $4.40Memory (4-bit): 193 GB → 452 GB+1GLM 5.2maxOpenwins 64%—$1.28Memory (4-bit): 193 GB → 452 GB+1
MiMo v2.6 FlashunstatedOpenwins 64%—$0.14 / $0.28Pass back reasoning_content+3GLM 5.3 FlashunstatedOpenwins 64%—$0.15 / $0.50Defaults to max · Self-hosted+4Run it yourself
estimateFits on✗ 8 GB (phone, laptop)✗ 16 GB✗ 24 GB GPU✗ 48 GB GPU✗ 80 GB GPU
LicenceApache-2.0free for commercial use
Benchmarks
all categoriesChat & Knowledge
Arena Text · overall1,349#124
Arena Text · Chinese1,397.5#115
Arena Text · English1,373.1#126
Arena Text · German1,370.7#92
Arena Text · instruction following1,332.2#124
Arena Text · Korean1,269.2#112
Arena Text · non-English1,327.3#128
Arena Text · Polish1,300.6#118
Arena Text · Russian1,330.8#129
Arena Text · Spanish1,371#111
Arena Text · overall · October 22, 20261,350.4#122
Arena Text · overall · October 15, 20261,350.3#119
Writing & Language
Arena Text · creative writing1,310#122
Vision & Documents
Arena Vision · overall1,176.4#68
FAQ
How much does Step3 cost?›
There is no per-unit API price: self-host only.
Is Step3 free to use?›
There is no free API. The weights are open, so you can run it yourself.
How good is Step3, and what is it best at?›
#126 of 205 in Chat & Knowledge, #115 of 197 in Writing & Language, #68 of 88 in Vision & Documents.
Is Step3 better than GPT OSS 120B?›
On Arena Text · English, Step3 scores 1,373.1 vs 1,374.7 — about the same.
When was Step3 released? Is there a newer version?›
Released 2025-07-28. It is the current version in its line.
Is Step3 open source?›
Yes, the weights are open (Apache-2.0 licence).
Can I run Step3 locally?›
The weights are open and can be self-hosted; hardware needs are not confirmed yet.
Can Step3 read images?›
It accepts text and images as input.
What is Step3?›
Step3 is an open-weights language model from stepfun-ai, released Jul 2025.
What are the best alternatives to Step3?›
GPT-6 Luna (better), GLM 5.3 (open weights).