Step 5 Preview: Specifications & Sources

Specifications and access

Specification	Value and source
Provider model ID	step-5-previewSource
Access	API platformSource
Input modalities	Text, image, videoSource
Output modalities	TextSource
Announced weight release	StepFun has announced open weights for October 15, 2026.Source
Reported context	1M tokens; no exact integer limit statedSource
Maximum output	64K tokensSource
Images per request	Up to 60 imagesSource

Measurements without matching peer values

These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.

Vals Index (Show measurement, test conditions, and source)

Source value: 49.35
Score: 49.35
Metric: Vals AI reported benchmark score
Unit: %
Category: aggregate
Direction: Higher is better
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-10-07
Methodology: Independent evaluation. Vals AI page-level settings: default provider StepFun, temperature 1, top-p 0.95, top-k default, maximum output 1,024,000 tokens. Vals says some benchmarks may use different providers or parameters; no per-benchmark override is specified on this page.
Context: Independent evaluator results. The page's ± values are reported error bars; Vals methodology defines these as standard errors of the mean (SEM), not 95% confidence intervals. The ranking fractions on the page are ranks, not sample counts.

Source Vals AI

Code Migration (Show measurement, test conditions, and source)

Source value: 38.98
Score: 38.98
Metric: Vals AI reported benchmark score
Unit: %
Category: coding
Direction: Higher is better
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-10-07
Methodology: Independent evaluation. Vals AI page-level settings: default provider StepFun, temperature 1, top-p 0.95, top-k default, maximum output 1,024,000 tokens. Vals says some benchmarks may use different providers or parameters; no per-benchmark override is specified on this page.
Context: Independent evaluator results. The page's ± values are reported error bars; Vals methodology defines these as standard errors of the mean (SEM), not 95% confidence intervals. The ranking fractions on the page are ranks, not sample counts.

Source Vals AI

EMB (Show measurement, test conditions, and source)

Source value: 57.12
Score: 57.12
Metric: Vals AI reported benchmark score
Unit: %
Category: reasoning
Direction: Higher is better
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-10-07
Methodology: Independent evaluation. Vals AI page-level settings: default provider StepFun, temperature 1, top-p 0.95, top-k default, maximum output 1,024,000 tokens. Vals says some benchmarks may use different providers or parameters; no per-benchmark override is specified on this page.
Context: Independent evaluator results. The page's ± values are reported error bars; Vals methodology defines these as standard errors of the mean (SEM), not 95% confidence intervals. The ranking fractions on the page are ranks, not sample counts.

Source Vals AI

Finance Agent (v2) (Show measurement, test conditions, and source)

Source value: 50.67
Score: 50.67
Metric: Vals AI reported benchmark score
Unit: %
Category: finance
Direction: Higher is better
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-10-07
Methodology: Independent evaluation. Vals AI page-level settings: default provider StepFun, temperature 1, top-p 0.95, top-k default, maximum output 1,024,000 tokens. Vals says some benchmarks may use different providers or parameters; no per-benchmark override is specified on this page.
Context: Independent evaluator results. The page's ± values are reported error bars; Vals methodology defines these as standard errors of the mean (SEM), not 95% confidence intervals. The ranking fractions on the page are ranks, not sample counts.

Source Vals AI

Legal Research Bench (Show measurement, test conditions, and source)

Source value: 40.38
Score: 40.38
Metric: Vals AI reported benchmark score
Unit: %
Category: legal
Direction: Higher is better
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-10-07
Methodology: Independent evaluation. Vals AI page-level settings: default provider StepFun, temperature 1, top-p 0.95, top-k default, maximum output 1,024,000 tokens. Vals says some benchmarks may use different providers or parameters; no per-benchmark override is specified on this page.
Context: Independent evaluator results. The page's ± values are reported error bars; Vals methodology defines these as standard errors of the mean (SEM), not 95% confidence intervals. The ranking fractions on the page are ranks, not sample counts.

Source Vals AI

MysteryMechanism (Show measurement, test conditions, and source)

Source value: 18.47
Score: 18.47
Metric: Vals AI reported benchmark score
Unit: %
Category: reasoning
Direction: Higher is better
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-10-07
Methodology: Independent evaluation. Vals AI page-level settings: default provider StepFun, temperature 1, top-p 0.95, top-k default, maximum output 1,024,000 tokens. Vals says some benchmarks may use different providers or parameters; no per-benchmark override is specified on this page.
Context: Independent evaluator results. The page's ± values are reported error bars; Vals methodology defines these as standard errors of the mean (SEM), not 95% confidence intervals. The ranking fractions on the page are ranks, not sample counts.

Source Vals AI

ProofBench v1.1 (Show measurement, test conditions, and source)

Source value: 42
Score: 42
Metric: Vals AI reported benchmark score
Unit: %
Benchmark version: v1.1
Category: reasoning
Direction: Higher is better
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-10-07
Methodology: Independent evaluation. Vals AI page-level settings: default provider StepFun, temperature 1, top-p 0.95, top-k default, maximum output 1,024,000 tokens. Vals says some benchmarks may use different providers or parameters; no per-benchmark override is specified on this page.
Context: Independent evaluator results. The page's ± values are reported error bars; Vals methodology defines these as standard errors of the mean (SEM), not 95% confidence intervals. The ranking fractions on the page are ranks, not sample counts.

Source Vals AI

Public Benefits Bench v1.1 (Show measurement, test conditions, and source)

Source value: 58.12
Score: 58.12
Metric: Vals AI reported benchmark score
Unit: %
Benchmark version: v1.1
Category: public-benefits
Direction: Higher is better
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-10-07
Methodology: Independent evaluation. Vals AI page-level settings: default provider StepFun, temperature 1, top-p 0.95, top-k default, maximum output 1,024,000 tokens. Vals says some benchmarks may use different providers or parameters; no per-benchmark override is specified on this page.
Context: Independent evaluator results. The page's ± values are reported error bars; Vals methodology defines these as standard errors of the mean (SEM), not 95% confidence intervals. The ranking fractions on the page are ranks, not sample counts.

Source Vals AI

Tax Agent Bench (Show measurement, test conditions, and source)

Source value: 59.54
Score: 59.54
Metric: Vals AI reported benchmark score
Unit: %
Category: finance
Direction: Higher is better
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-10-07
Methodology: Independent evaluation. Vals AI page-level settings: default provider StepFun, temperature 1, top-p 0.95, top-k default, maximum output 1,024,000 tokens. Vals says some benchmarks may use different providers or parameters; no per-benchmark override is specified on this page.
Context: Independent evaluator results. The page's ± values are reported error bars; Vals methodology defines these as standard errors of the mean (SEM), not 95% confidence intervals. The ranking fractions on the page are ranks, not sample counts.

Source Vals AI

Vibe Code Bench v1.1 (Show measurement, test conditions, and source)

Source value: 70.4
Score: 70.4
Metric: Vals AI reported benchmark score
Unit: %
Benchmark version: v1.1
Category: coding
Direction: Higher is better
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-10-07
Methodology: Independent evaluation. Vals AI page-level settings: default provider StepFun, temperature 1, top-p 0.95, top-k default, maximum output 1,024,000 tokens. Vals says some benchmarks may use different providers or parameters; no per-benchmark override is specified on this page. The exact stepfun/step-5-preview overall row on the separately read benchmark page identifies the agent harness as OpenHands (https://www.vals.ai/benchmarks/vibe-code).
Context: Independent evaluator results. The page's ± values are reported error bars; Vals methodology defines these as standard errors of the mean (SEM), not 95% confidence intervals. The ranking fractions on the page are ranks, not sample counts.

Source Vals AI

BioMysteryBench (Show measurement, test conditions, and source)

Source value: 67.41
Score: 67.41
Metric: Vals AI reported benchmark score
Unit: %
Category: science
Direction: Higher is better
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-10-07
Methodology: Independent evaluation. Vals AI page-level settings: default provider StepFun, temperature 1, top-p 0.95, top-k default, maximum output 1,024,000 tokens. Vals says some benchmarks may use different providers or parameters; no per-benchmark override is specified on this page.
Context: Independent evaluator results. The page's ± values are reported error bars; Vals methodology defines these as standard errors of the mean (SEM), not 95% confidence intervals. The ranking fractions on the page are ranks, not sample counts.

Source Vals AI

Harvey's Legal Agent Benchmark (Show measurement, test conditions, and source)

Source value: 9.17
Score: 9.17
Metric: Vals AI reported benchmark score
Unit: %
Category: legal
Direction: Higher is better
Source type: independent-evaluator
Evaluator: Vals AI
Status: active
Retrieved at: 2026-10-07
Methodology: Independent evaluation. Vals AI page-level settings: default provider StepFun, temperature 1, top-p 0.95, top-k default, maximum output 1,024,000 tokens. Vals says some benchmarks may use different providers or parameters; no per-benchmark override is specified on this page.
Context: Independent evaluator results. The page's ± values are reported error bars; Vals methodology defines these as standard errors of the mean (SEM), not 95% confidence intervals. The ranking fractions on the page are ranks, not sample counts.

Source Vals AI

Page 1 of 7

Our tests

Each test shows the latest attempt, with no subsequent fixes to the generated code. Earlier runs remain available in the test details. These results do not contribute to an overall score.

This is the lowest documented setting for each provider. It does not make compute comparable across models.

Pelican on a bicycle

A community classic for free-form SVG drawing.

Step 5 Preview

Output withheld for safety

Reasoning (requested): low

Current attempt

Run date: Oct 11, 2026
API provider: StepFun
Requested API model: stepfun/step-5-preview
API model reported in response: stepfun/step-5-preview
Reasoning setting: low
Pinned endpoint: StepFun | stepfun/step-5-preview-20261008
Temperature: Provider default
Token limit including reasoning: 8,192 tokens
Reasoning tokens: 455
Test protocol: community-visual-low-v1
Technical checks: Failed
Response time: 46.1 s

Not yet visually reviewed. A successful recording does not confirm correct physics or full compliance with the prompt.

Prompt and test conditions

Generate an SVG of a pelican riding a bicycle

Token limit including reasoning: 8,192 tokens

Task origin (Simon Willison)

Sources and data date

Type	Evidence and data date
Input	Documented modalities. Retrieved October 7, 2026. Source: Step 5 Preview API model documentation · Step 5 Preview API model documentation (retrieved October 7, 2026) · Step 5 Preview API model documentation · Editorial description reviewed October 7, 2026
Output	Documented modalities. Retrieved October 7, 2026. Source: Step 5 Preview API model documentation · Step 5 Preview API model documentation (retrieved October 7, 2026) · Step 5 Preview API model documentation · Editorial description reviewed October 7, 2026
Research status	Research date October 7, 2026. 20 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication.
Unresolved	first_party_benchmarks Original wording The independent evaluators checked here publish numerical results. StepFun's official model page was not readable, so no additional vendor benchmark was included.
Additional source	StepFun official website (retrieved October 7, 2026) · StepFun official website · Editorial description reviewed October 7, 2026
Additional source	Vals AI (retrieved October 7, 2026)
Additional source	Artificial Analysis (retrieved October 7, 2026)
Additional source	Vals AI (retrieved October 7, 2026)
Additional source	Vals AI (retrieved October 7, 2026)
Additional source	Vals AI (retrieved October 7, 2026)
Additional source	Vals AI (retrieved October 7, 2026)
Additional source	Vals AI (retrieved October 7, 2026)
Additional source	Vals AI (retrieved October 7, 2026)
Additional source	Vals AI (retrieved October 7, 2026)
Additional source	Vals AI (retrieved October 7, 2026)
Additional source	Vals AI (retrieved October 7, 2026)
Additional source	Vals AI (retrieved October 7, 2026)
Additional source	Vals AI (retrieved October 7, 2026)
Additional source	Vals AI (retrieved October 7, 2026)
Additional source	Vals AI (retrieved October 7, 2026)
Additional source	Vals AI (retrieved October 7, 2026)
Additional source	Vals AI (retrieved October 7, 2026)
Additional source	Model metadata (retrieved October 7, 2026)