Data as of November 5, 2026 · How the score is built
Released Sep 20, 20261M contextStepFun Step 5 Preview model card
Step 5 Preview
Decision readingStep 5 Preview scores 56.8 out of 100 and ranks #60 of 216. This profile shows 24 source-displayable benchmark rows; its strongest eligible category is Multimodal & Grounded at #32. API pricing is $1 input and $2.7 output per million tokens, with cached input at $0.05.
Released Sep 20, 2026 — see all recent releases
Decision snapshot
Each value carries a field reference instead of floating alone. Markers compare this model with the current ranked and priced catalog; they are not absolute quality thresholds.
Capability
56.8/100
field median 50.1#60 of 216 ranked models
Public
#60of 216
Verified —
Price
$1input / $2.70 output
input median $0.65cached $0.050 · blended $1.85
Speed
181tok/s
field median 92 tok/sFirst token 11.96 s
Context
1Mtokens
field median 256,000Maximum output length is tracked separately
Strongest published evidence
Multimodal & Grounded ranks #32. Particularly strong for screenshots, documents, charts, and grounded multimodal workflows.
Validate before choosing
24 published rows leave some tracked benchmark slots empty. Coding is its lowest eligible category at #47.
Source-linked · 24 displayable benchmark rows
Category score record
Scores and ranks appear only where published evidence can be displayed. The table keeps the score, weight, cohort, and evidence state together.
| Category | Score | Rank | Percentile | Weight | Benchmarks | Evidence |
|---|---|---|---|---|---|---|
| AgenticRank #42 of 135Percentile 69thWeight 22%13 benchmarksVerified | 55.3 | 13 benchmarks | Verified | |||
| CodingRank #47 of 156Percentile 70thWeight 20%6 benchmarksVerified | 48.8 | 6 benchmarks | Verified | |||
| ReasoningRank Not rankedWeight 17%1 benchmarkVerified | 82.5 | 1 benchmark | Verified | |||
| MultimodalRank #32 of 58Percentile 46thWeight 12%2 benchmarksVerified | 73.3 | 2 benchmarks | Verified | |||
| KnowledgeRank #44 of 177Percentile 76thWeight 12%2 benchmarksVerified | 61.0 | 2 benchmarks | Verified | |||
| MultilingualWeight 7%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured | |||
| Inst. FollowingWeight 5%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured | |||
| MathWeight 5%0 benchmarksNot measured | Not measured | 0 benchmarks | Not measured |
24 of 802 tracked benchmark slots have displayable evidence · bars run 0–100
Coverage detailsHow much of this is verified
Coverage is split by category so a strong number never hides a thin evidence base. Verified means the row is tied to a published source; provisional rows remain visible but separate.
- Agentic13/13 verified
- Coding6/6 verified
- Reasoning1/1 verified
- Multimodal2/2 verified
- Knowledge2/2 verified
- MultilingualNot measured
- Inst. FollowingNot measured
- MathNot measured
Capability shape
Each axis shows percentile within that category’s eligible cohort. The comparison outline is the median of the six nearest public-score peers; a collapsed vertex means the category is not rank-eligible.
Step 5 Preview category percentile values
- Agentic69th percentile
- Coding70th percentile
- ReasoningNot eligible
- Multimodal46th percentile
- Knowledge76th percentile
- MultilingualNot eligible
- Instruction followingNot eligible
- MathNot eligible
The dashed outline is median of 6 nearest peers.
Eligible category ranks
- Agentic#42/135
- Coding#47/156
- ReasoningNot ranked
- Multimodal#32/58
- Knowledge#44/177
- MultilingualNot ranked
- Inst. FollowingNot ranked
- MathNot ranked
Benchmark ledger
Coding opens by default. The marker compares each value with the best source-verified result in the catalog; provisional leaders do not set the reference. Expand the remaining categories for every published row.
Coding6 rows
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| DeepSWE | Score67.7% | Versus best verified row Best verified: Muse Wonder Full · 79.0% | Gap11.3 behind | Weight15% ref. weight | Provider exact |
| SciCodeScientific Code Benchmark | Score58.9% | Versus best verified row Best verified: Sakana Fugu · 60.1% | Gap1.2 behind | Weight10% ref. weight | Provider exact |
| Terminal-Bench 2.1 | Score85.0% | Versus best verified row Best verified: SWE-2 · 92.8% | Gap7.8 behind | WeightScored in Agentic | Provider exact |
| ProgramBenchProgramBench: Can Language Models Rebuild Programs From Scratch? | Score80.5% | Versus best verified row Best verified: Claude Opus 5 · 93.0% | Gap12.5 behind | WeightDisplay only | Provider exact |
| sweMarathon | Score72.7% | Versus best verified row Best verified: Step 5 Preview · 72.7% | GapBest verified | WeightDisplay only | Provider exact |
| MLS-Bench Lite | Score40.5% | Versus best verified row Best verified: Kimi K3 · 48.3% | Gap7.8 behind | WeightDisplay only | Provider exact |
Agentic13 rows
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| Terminal-Bench 2.1 | Score85.0% | Versus best verified row Best verified: SWE-2 · 92.8% | Gap7.8 behind | Weight8% ref. weight | Provider exact |
| Terminal-Bench 4.0 | Score33.30% | Versus best verified row Best verified: Claude Sonnet 5.5 · 70.60% | Gap37.3 behind | Weight8% ref. weight | Provider exact |
| BrowseComp | Score88.7% | Versus best verified row Best verified: Atria Dawn Preview · 92.5% | Gap3.8 behind | Weight8% ref. weight | Provider exact |
| AutomationBench | Score44.0% | Versus best verified row Best verified: MiniMax M3.1 Flash · 62.2% | Gap18.2 behind | Weight6% ref. weight | Provider exact |
| JobBench | Score59.0% | Versus best verified row Best verified: Muse Wonder Full · 75.3% | Gap16.3 behind | Weight6% ref. weight | Provider exact |
| MCP Atlas | Score85.6% | Versus best verified row Best verified: Muse Spark Open · 89.2% | Gap3.6 behind | Weight4% ref. weight | Provider exact |
| Toolathlon-Verified | Score74.1% | Versus best verified row Best verified: Claude Opus 5 · 80.6% | Gap6.5 behind | Weight3% ref. weight | Provider exact |
| Agents' Last Exam | Score29.5% | Versus best verified row Best verified: GPT-6.1 Terra · 60.0% | Gap30.5 behind | Weight3% ref. weight | Provider exact |
| CyberGym | Score84.7% | Versus best verified row Best verified: MiMo-V2.6-Flash · 95.1% | Gap10.4 behind | WeightDisplay only | Provider exact |
| APEX-Agents | Score37.8% | Versus best verified row Best verified: Hy4 · 75.2% | Gap37.4 behind | WeightDisplay only | Provider exact |
| DRACOData Research and Analysis with Complex Operations | Score83.3% | Versus best verified row Best verified: Claude Opus 5 · 88.6% | Gap5.3 behind | WeightDisplay only | Provider exact |
| Hermes IndexHermes Index: agentic capability and cost per task inside Hermes Agent | Score32.8% | Versus best verified row Best verified: Claude Opus 5.5 · 63.3% | Gap30.5 behind | WeightDisplay only | Benchmark exact |
| Hermes BenchHermes Bench: 150 agent tasks graded on the files and state the agent leaves behind | Score69.2% | Versus best verified row Best verified: Claude Opus 5.5 · 76.7% | Gap7.5 behind | WeightDisplay only | Benchmark exact |
Reasoning1 row
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| CritPtCritical Physics Tasks | Score20.9% | Versus best verified row Best verified: Muse Wonder Full · 32.9% | Gap12 behind | WeightFeeds overall | Provider exact |
Multimodal2 rows
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| MMMU-ProMassive Multi-discipline Multimodal Understanding Pro | Score76% | Versus best verified row Best verified: GPT-6.1 Terra · 85.5% | Gap9.5 behind | WeightWeighted 40% | Provider exact |
| OfficeQA Pro | Score60.3% | Versus best verified row Best verified: Claude Opus 5.5 · 67.7% | Gap7.4 behind | WeightWeighted 25% | Provider exact |
Knowledge2 rows
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| HLEHumanity's Last Exam | Score46.5% | Versus best verified row Best verified: Kimi K3.1 Pro · 70% | Gap23.5 behind | Weight44% ref. weight | Provider exact |
| GPQA-DGPQA Diamond | Score93.5% | Versus best verified row Best verified: GPT-6 Astra · 96.0% | Gap2.5 behind | WeightDisplay only | Provider exact |
Bars run 0–100; the dark tick marks the best source-verified value
All 24 rowsWhat it costs to get this score
Published API price against the public score. The x-axis uses a log scale; the dashed path marks models that are not beaten by a cheaper, higher-scoring option. Price uses average of published input and output rates.
Step 5 Preview · 56.8 score · $1.85 blended per million tokens
The chart opens on the current model. Scroll horizontally to inspect the full price axis.
Horizontal: blended price per million tokens, log scale · Vertical: public score
Lineage
The sequence follows explicit supersedes links. Each score is estimated for that model; a relative can inform a sparse estimate but never sets a floor, so a newer release can score below an earlier one. Scores and prices remain blank when the corresponding public row or first-party rate is unavailable.
preview · Preview
Spec sheet
Each documented value carries its source. Missing fields stay visible as not sourced or not published, rather than disappearing from the page.
- API model ID
- step-5-previewStepFun Step 5 Preview model documentation
- Context window
- 1MStepFun Step 5 Preview model documentation
- Maximum output
- 64K tokensStepFun Step 5 Preview model documentation
- Knowledge cutoff
- Not sourced yet
- Input modalities
- text, image, videoStepFun Step 5 Preview model documentation
- Output modalities
- textStepFun Step 5 Preview model documentation
- Parameters
- Not sourced yet
- Availability
- StepFun Open Platform API · Hugging Face · self-hosted inferenceStepFun Step 5 Preview model documentation
- Cloud regions
- Not tracked yet
- Lifecycle
- previewStepFun Step 5 Preview model documentation
- API capabilities
- Streaming output · tool calling · JSON Mode / JSON Schema · prompt cachingStepFun Step 5 Preview model documentation
- Prompt caching
- Published at $0.050 per million cached input tokensStepFun pricing and rate limits
- Self-host
- Step 5 Preview needs ~640GB VRAM (8× NVIDIA H100 (80GB)).
- Rate limits
- Not tracked yet
How to read this profile
The visual layer above carries the decisions. These notes preserve the model, ranking, coverage, and family context behind the numbers.
Step 5 Preview ranks #60 of 216 on the public leaderboard with a score of 56.81/100. It does not yet have enough sourced coverage for a verified position.
Step 5 Preview is a open weight model with a 1M context window. It uses an explicit reasoning mode, which can improve complex problem solving while adding latency and token use.
Available through StepFun's Open Platform API as `step-5-preview` and as downloadable FP8 and BF16 weights. The self-hosted model card documents up to 1M tokens for input and output combined; the hosted API has its own output limit. Weights use StepFun Model License 1.0, with commercial-use conditions.
StepFun publishes a 600B-total, 27B-active MoE with up to 1M context and FP8 and BF16 weights under StepFun Model License 1.0. Stored launch benchmarks use High reasoning effort and retain their original sources and protocol exclusions. The custom license includes commercial-use conditions. Weight publication does not establish new benchmark results.
24 of 802 tracked benchmark slots currently have displayable evidence. Missing categories stay blank.
Its strongest eligible category is Multimodal & Grounded at #32, while its lowest eligible position is Coding at #47. particularly strong for screenshots, documents, charts, and grounded multimodal workflows.
Last updated November 5, 2026. Runtime fields remain blank until a sourced snapshot exists.
Deployment options
Self-host and provider-specific paths stay separate from benchmark evidence so operating constraints are visible before a score becomes the whole decision.
Where to run Step 5 Preview by API
3 hosts publish a per-token price for it. DeepInfra has the lowest output rate, $2.70 per million tokens. Prices read from each host’s own pages on October 20, 2026.
| Host | Input /1M | Output /1M | Cached /1M | Context |
|---|---|---|---|---|
| DeepInfra | $1.00 | $2.70 | $0.10 | 1.05M |
| Novita AI | $1.00 | $2.70 | $0.10 | 1M |
| OpenRouter | $1.00 | $2.70 | $0.10 | 1.05M |
Self-host vs API cost
Estimates at 50,000 req/day · 1000 tokens/req average.
Questions
How does Step 5 Preview perform overall in AI benchmarks?
Step 5 Preview ranks #60 out of 216 models on the public BenchAlign leaderboard, with a score of 56.81/100. Its evidence status is Estimated, and this profile shows 24 source-displayable benchmark rows. The label describes evidence depth, not a provider quality claim; inspect category rows before choosing a workload.
Is Step 5 Preview good for knowledge and understanding?
Step 5 Preview ranks #44 out of 177 eligible models for knowledge and understanding, with a public category score of 61/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is Step 5 Preview good for coding and programming?
Step 5 Preview ranks #47 out of 156 eligible models for coding and programming, with a public category score of 48.8/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is Step 5 Preview good for reasoning and logic?
Step 5 Preview has source-displayable benchmark coverage for reasoning and logic, but the public category table does not assign it a rank there. The individual rows remain available for inspection. A missing category position means the evidence threshold was not met; it does not convert the model's unmeasured work into a zero.
Is Step 5 Preview good for agentic tool use and computer tasks?
Step 5 Preview ranks #42 out of 135 eligible models for agentic tool use and computer tasks, with a public category score of 55.3/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is Step 5 Preview good for multimodal and grounded tasks?
Step 5 Preview ranks #32 out of 58 eligible models for multimodal and grounded tasks, with a public category score of 73.3/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is Step 5 Preview open source?
Step 5 Preview is an open-weight model from StepFun. Its weights can be downloaded for local or hosted deployment, subject to the published license. Open weight does not automatically mean open source: training data and training code may remain private, and commercial restrictions can still apply.
Does Step 5 Preview have full benchmark coverage on BenchLM?
No. Step 5 Preview currently has 42 source-displayable rows. The profile exposes published, non-generated evidence and leaves missing categories blank until an exact evaluation is available. Coverage describes how much was measured; it is not a penalty added to an individual benchmark result.
What is the context window size of Step 5 Preview?
Step 5 Preview has a documented context window of 1M. That figure is the maximum combined prompt and retained-conversation space reported for this exact model; it is not the maximum output length. The profile keeps output limits separate because providers often publish those limits independently.