Step3 VL 10B
Jan 2026
Step3 VL 10BCompare inon
Benchmarks of Step3 VL 10B11
BenchmarkByScoreRank
Overall1
Knowledge4
Science2
Coding1
Multimodal1
Legacy · saturated or no longer reported2
No benchmark matches
—per task?
self-host onlyprice
—output speed?
66Kcontext
Ranking?
Alternatives?
ModelScoreCostPer taskWhat changes
BetterGPT-6 Lunalow+87%—$0.005Function calling only with none · Chat Completions+3IOpenK2 Horizon 7Bthinking+68%—$0.30 / $1.20No image inputBetterGPT-5.6 Lunalow+83%—$0.01
BetterMiMo v2.5thinkingOpen×2.5—$0.019Memory (4-bit): 5.1 GB → 186 GB
BetterClaude Haiku 5.5low×2.5—$0.024Remove sampling parameters+5ModelScoreCostPer taskWhat changes
Nothing here yet.
ModelScoreCostPer taskWhat changes
GPT-6 Lunalow+87%—$0.005Function calling only with none · Chat Completions+3GPT-5.6 Lunalow+83%—$0.01MiMo v2.5thinkingOpen×2.5—$0.019Memory (4-bit): 5.1 GB → 186 GBClaude Haiku 5.5low×2.5—$0.024Remove sampling parameters+5
Mercury 2high+58%—$0.039No image inputModelScoreCostPer taskWhat changes
IK2 Horizon 7BthinkingOpen+68%—$0.30 / $1.20No image input
Gemma 4 12BthinkingOpen+44%—$0.10 / $0.30Memory (4-bit): 5.1 GB → 6 GB
Qwen 3.5 9BthinkingOpen+38%—$0.21IK2 Horizon 3.7BthinkingOpen+28%—$0.30 / $1.20No image inputSApriel v1.5 15B ThinkerOpen+12%——Memory (4-bit): 5.1 GB → 7.4 GBModelScoreCostPer taskWhat changes
Step 5 PreviewthinkingOpen×4.3—$1.03Memory (4-bit): 5.1 GB → 363 GB+3Step 3.7 FlashhighOpen+97%—$0.20 / $1.15Memory (4-bit): 5.1 GB → 121 GBStep 3.5 Flash 2603n/a—$0.10 / $0.30No image inputStep 1o Turbo Visionn/a—$0.36 / $1.15Context: not verifiedAll StepFun models →Run it yourself
estimateFits on✓ 8 GB (phone, laptop) · compressed✓ 16 GB · 8-bit✓ 24 GB GPU · full quality✓ 48 GB GPU · full quality✓ 80 GB GPU · full quality
LicenceApache-2.0free for commercial use
Benchmarks
all categoriesChat & Knowledge
Humanity's Last Exam10.8#180
AA Omniscience Index-59#252
CritPt (physics)0#180
AA Intelligence Index 4.3.2≈7.7#257
AA Omniscience · Accuracy13.3#258
GPQA Diamond69#191
Coding
Agents & Tool Use
τ²-bench · AA16.1#207
Writing & Language
IFBench (instructions)50.2#116
Vision & Documents
MMMU-Pro (vision)64#88
FAQ
How much does Step3 VL 10B cost?›
There is no per-unit API price: self-host only.
Is Step3 VL 10B free to use?›
There is no free API. The weights are open, so you can run it yourself.
How good is Step3 VL 10B, and what is it best at?›
#179 of 334 in Chat & Knowledge, #156 of 239 in Coding, #204 of 242 in Agents & Tool Use, #115 of 245 in Writing & Language, #86 of 124 in Vision & Documents.
Is Step3 VL 10B better than DiffusionGemma 26B A4B?›
On Humanity's Last Exam, Step3 VL 10B scores 10.8 vs 10.8 — about the same.
What is the context window of Step3 VL 10B?›
66K tokens.
When was Step3 VL 10B released? Is there a newer version?›
Released 2026-01-20. It is the current version in its line.
Is Step3 VL 10B open source?›
Yes, the weights are open (Apache-2.0 licence).
Can I run Step3 VL 10B locally?›
The weights are open and can be self-hosted; hardware needs are not confirmed yet.
Can Step3 VL 10B read images?›
It accepts text and images as input.
What is Step3 VL 10B?›
Step3 VL 10B is an open-weights language model from StepFun, released Jan 2026.
What are the best alternatives to Step3 VL 10B?›
GPT-6 Luna (better), K2 Horizon 7B (open weights).