Step 5 Preview
- Input
- $1.00/M tok
- Output
- $2.70/M tok
- Blended
- $1.425/M tok · 3:1
- Speed
- 85tok/s
- Context
- 1Mtokens
- Cost per task
- ~$1.1/taskest.
Open weights since 2026-10-15 (FP8, BF16 and NVFP4 repos at huggingface.co/stepfun-ai/Step-5-Preview, StepFun Model License 1.0); 600B total / 27B active MoE. Price is StepFun's API price. Writing: no LMArena/EQ-Bench entry or lab-reported writing benchmark as of 2026-09-24, so its Writing entry is fully estimated from its other categories (flagged “Estimated from other categories”).
Rankings
Coding
Verified#27 / 137rank by quality68.0±4.5quality / 1004/7 benchmarks measuredWriting & chat
Verified#34 / 162rank by quality67.5±6.0quality / 1002/5 benchmarks measuredVision
Verified#32 / 98rank by quality65.0±6.3quality / 1002/6 benchmarks measuredHard reasoning
Mixed#18 / 218rank by quality77.9±4.0quality / 1003/4 benchmarks measuredAgents
Mixed#22 / 187rank by quality74.1±6.3quality / 1003/7 benchmarks measured
Self-hostingEstimated
How we estimate →What it would cost to run these open weights yourself on rented GPUs, compared with the API price above.
- Hardware
- 8× B200 180GB
- FP8 weights
- Throughput
- ~28,552 tok/s
- many requests batched
- Price
- $0.20–1.76 /M tok
- busy → light use
- On your own machine
- Mac Studio M5 Ultra 512GB (int4, ~62 tok/s single-stream)
- single consumer GPU or Mac
Assumptions (5)
- FP8 weights (600B params, 27B active per token) + 40% KV-cache headroom ≈ 840 GB
- 8× B200 180GB at $5.98–$14.24/GPU-hour on-demand (2026-09-24)
- ~28,552 output tok/s aggregate at batch 817 (bandwidth-bound); MoE compute scales with active params
- Blended 3:1 input:output; prefill ~315,000 tok/s
- Low = 75% utilization at the low GPU price; high = 20% at the high price
Benchmark scores
~ italic, dashed = no published score yet, estimated from related benchmarks.
Coding
SWE-bench VerifiedEstimated
~#2 of 21 (estimated)~87.4%
No published score — estimated from related benchmarks
SWE-bench Pro (Public)Estimated
~#6 of 24 (estimated)~69.7%
- #1Claude Fable 5.293.0%
- #2Claude Opus 5.589.9%
- #5Claude Opus 579.2%
- ~#6Step 5 PreviewEstimated69.7%
- #6Intern-S2-397B68.5%
No published score — estimated from related benchmarks
DeepSWE v1.1Verified
#27 of 4561.9%
- #1Gemini 4 Argon78.8%
- #2DeepSeek V4.1 Flash 101777.0%
- #26Claude Opus 562.5%
- #27Step 5 Preview61.9%
- #28Mistral Large 461.7%
artificialanalysis.ai · 2026-10-10
Terminal-Bench 4.0Verified
#19 of 10133.3%
- #1Kimi K3.1 Pro67.2%
- #2Claude Sonnet 5.563.6%
- #19Muse Spark 1.333.3%
- #19Step 5 Preview33.3%
- #22Claude Haiku 5.532.8%
artificialanalysis.ai · 2026-09-24
LiveCodeBenchEstimated
~#10 of 28 (estimated)~86.0%
- #1Qwen3.8-Omni-Flash92.6%
- #2Qwen3.8 Flash-Next91.9%
- #9Nemotron 3.5 Super VL NVFP487.4%
- ~#10Step 5 PreviewEstimated86.0%
- #10Ternary Bonsai 27B82.8%
No published score — estimated from related benchmarks
SciCodeVerified
#12 of 11858.9%
- #1Claude Fable 5.267.7%
- #2Claude Opus 5.566.9%
- #11GLM-5.359.0%
- #12Step 5 Preview58.9%
- #13Muse Spark 1.358.8%
artificialanalysis.ai · 2026-09-24
Writing & chat
IFBenchEstimated
~#26 of 140 (estimated)~74.3%
- #1MiniMax-M382.9%
- #2Qwen3.8 2.4T A95B82.8%
- #25LFM2.5-24B-A2B74.7%
- ~#26Step 5 PreviewEstimated74.3%
- #26Ternary Bonsai 2 27B74.0%
No published score — estimated from related benchmarks
LMArena Text, Non-English (Elo)Verified
#29 of 561433 Elo
- #1Gemini 4 ArgonFew votes1512
- #2Claude Opus 5.51497
- #28Qwen3.5 397B A17B1434
- #29Step 5 Preview1433
- #30Grok 4.71432
lmarena.ai · 2026-10-08
EQ-Bench Creative Writing v3 (Elo)Estimated
~#9 of 26 (estimated)~1926 Elo
- #1GPT-6 Astra2173
- #2Claude Fable 5.12162
- #8Grok 4.72007
- ~#9Step 5 PreviewEstimated1926
- #9Muse Spark 1.31906
No published score — estimated from related benchmarks
Vision
MMMU-ProVerified
#31 of 8776.4%
- #1Claude Opus 5.587.7%
- #2GPT-6 Astra86.9%
- #30Grok Build 0.1 061676.5%
- #31Step 5 Preview76.4%
- #32GLM-5.3 Flash NVFP476.3%
artificialanalysis.ai · 2026-09-24
CharXiv ReasoningEstimated
~#6 of 12 (estimated)~81.3%
- #1Gemini 3.8 Flash86.2%
- #2Kimi K384.8%
- #5MiniMax-M3.1-Flash82.5%
- ~#6Step 5 PreviewEstimated81.3%
- #6Nemotron 3.5 Super VL80.8%
No published score — estimated from related benchmarks
OmniDocBench v1.5Estimated
~#4 of 8 (estimated)~87.9%
- #1MiniMax-M391.6%
- #2Kimi K391.1%
- #2Qwen3.8 27B91.1%
- ~#4Step 5 PreviewEstimated87.9%
- #4Gemini 3.1 Pro85.3%
No published score — estimated from related benchmarks
GDP.pdf (AA)Verified
#25 of 4714.8%
- #1Muse Wonder Full32.6%
- #2GPT-6 Astra31.0%
- #24GLM-5.3 Flash15.4%
- #25Step 5 Preview14.8%
- #26Gemini 3.5 Flash-Lite13.6%
artificialanalysis.ai · 2026-09-24
ChartographyEstimated
~#16 of 25 (estimated)~26.1%
- #1Claude Fable 5.273.4%
- #2GPT-6 Astra71.0%
- #15Gemini 3.1 Pro26.1%
- ~#16Step 5 PreviewEstimated26.1%
- #16Grok 4.620.5%
No published score — estimated from related benchmarks
Hard reasoning
GPQA DiamondLab-reported
#5 of 19293.5%
- #1GPT-6 Astra96.1%
- #2Gemini 3.8 Flash95.3%
- #5Qwen3.8 2.4T A95B93.5%
- #5Step 5 Preview93.5%
- #11Claude Opus 593.2%
stepfun.com · 2026-09-20
AIME (latest)Estimated
~#3 of 34 (estimated)~97.3%
No published score — estimated from related benchmarks
Agents
τ²-bench (Telecom)Estimated
~#3 of 120 (estimated)~98.0%
- #1Step 3.7 Flash98.5%
- #2DeepSeek-V4-Pro-0813-nvfp4-DSpark98.3%
- ~#3Step 5 PreviewEstimated98.0%
- #3Gemini 3.1 Pro95.6%
No published score — estimated from related benchmarks
OSWorld-VerifiedEstimated
~#5 of 15 (estimated)~81.8%
- #1Kimi K384.8%
- #2Qwen3.8 27B84.3%
- #4MiMo-V2.6-Pro82.0%
- ~#5Step 5 PreviewEstimated81.8%
- #5Claude Sonnet 581.2%
No published score — estimated from related benchmarks
OSWorld 2.0Estimated
~#3 of 5 (estimated)~54.1%
No published score — estimated from related benchmarks
BrowseCompLab-reported
#6 of 2188.7%
- #1Atria Dawn Preview92.5%
- #2GPT-6 Astra91.5%
- #5Nex-N2.5-Pro89.7%
- #6Step 5 Preview88.7%
- #7GPT-5.6 Terra87.5%
stepfun.com · 2026-09-20
GDPval-AA (Elo)Verified
#18 of 911584 Elo
- #1Claude Opus 5.51866
- #2Muse Wonder Full1857
- #17GPT-6.1 Sol1592
- #18Step 5 Preview1584
- #19GPT-6 Astra1574
artificialanalysis.ai · 2026-10-10
AutomationBench-AAVerified
#26 of 8451.0%
- #1Claude Fable 5.276.4%
- #2Muse Wonder Full71.3%
- #25GPT-6 Luna53.2%
- #26Step 5 Preview51.0%
- #27Qwen3.8 27B48.2%
artificialanalysis.ai · 2026-09-24
τ-Bench Banking (AA)Estimated
~#17 of 91 (estimated)~38.2%
- #1Muse Spark 1.350.5%
- #2GLM-5.350.3%
- #16GPT-5.539.0%
- ~#17Step 5 PreviewEstimated38.2%
- #17Claude Sonnet 537.3%
No published score — estimated from related benchmarks