Models / Step 5 Preview

StepFun

Step 5 Preview

Early evidence, 11 results from 1 source, on 2 use cases. Step 5 Preview is strongest at writing; there is too little evidence yet to recommend it either way.

Results 19 Nov 2026 · 11 results · 1 source

Price per million tokens
$1.00 in · $2.70 out
Typical host on OpenRouter, 10 Nov 2026 · compare prices
Speed
Not measured
Intelligence Index
Not measured
Spring Prompt overall
Not ranked yet
needs our benchmarks and two groups of results
Developer
StepFun, based in China
Weights published
Released
18 Sep 2026
1,000,000 tokens of context

Where it stands among the models you could choose

Available models grouped by price, intelligence, speed and developer location, plus models with published weights. Retired models are excluded.

AmongIntelligence IndexSpring Prompt overallPrice, blendedSpeed
Models available today – – $1.43 blended, 61st percentilecheapest: Mistral Nemo –
Models from developers based in China – – $1.43 blended, 82nd percentilecheapest: Qwen3.5-9B –
Open-weights models – – $1.43 blended, 86th percentilecheapest: Mistral Nemo –

The name under each rank is the leader of that group; green is the top quarter of the group and red the bottom quarter, by rank. Price is shown as a percentile: the share of the group that costs less, so lower is cheaper. "Sets this group" marks the measure the group is defined by. Spring Prompt overall is our score out of 100 across every source (how it works). Price is the developer's list price (or the typical OpenRouter host where we haven't read one) for three input tokens to one output; intelligence and speed are from Artificial Analysis (data sourced from Artificial Analysis); "similar intelligence" means within 4 points on its Intelligence Index. Developer locations are where each company is based, not where a model is served. Groups under 3 models are not ranked.

How good is it, and for what?

Each use case ranks the available models its benchmarks measured, then those priced $1–3 per million tokens. The bar shows its percentile in that field, best to the right. Open a row for the results behind it.

Product listingsTurning a sparse product feed and photos into listings that can go live Not measured
  • CatalogBench: has not measured this model.
Decks from an analysisTurning a finished analysis into a deck you could present as it is Not measured
  • DeckBench: has not measured this model.
User surveysPlanning a user survey and reading its results without being misled Not measured
  • SurveyBench: has not measured this model.
Marketing planningPlanning a year of ad spend without overspending Not measured
  • ROASBench: has not measured this model.
Agents and tool useMulti-step tasks with tools: support desks, coding agents, function calls PendingReleased after these sources last updated
  • tau2-bench: pending. Its latest results are dated 26 May 2026, before this model was released.
  • Berkeley Function Calling Leaderboard (BFCL) V4: pending. Its latest results are dated 16 Dec 2025, before this model was released.
  • OpenHands Index: pending. Its latest results are dated 30 Jun 2026, before this model was released.
  • Microsoft STATE-Bench: pending. Its latest results are dated 29 May 2026, before this model was released.
  • Artificial Analysis: has not measured this model.
  • Vending-Bench 2: has not measured this model.
Professional workReal tasks from banking, consulting and law, business documents and freelance projects Not measured
  • APEX-Agents: has not measured this model.
  • GDP.pdf: has not measured this model.
  • Remote Labor Index: has not measured this model.
Reasoning and knowledgeHard questions across science, maths and general knowledge Not measured
  • Artificial Analysis: has not measured this model.
WritingWhat people prefer in blind comparisons, and judged writing quality 54th of 169 available · 14th of 28 at its price
BenchmarkStep 5 PreviewRank among availableBest available
Arena (formerly LMArena)Arena Text: overall · reported 1,454 49th of 153 Gemini 4 Argon 1,525
  • UGI Leaderboard: has not measured this model.
Sticking to the factsSummarising without inventing things, and factual answers 54th of 139 available · 9th of 26 at its price
BenchmarkStep 5 PreviewRank among availableBest available
Arena (formerly LMArena)Arena Text factuality: overall · reported 1,452 45th of 109 Gemini 4 Argon 1,511
  • Vectara Hallucination Leaderboard: has not measured this model.
  • SimpleQA Verified (Epoch AI): has not measured this model.
Speed under pressureGood decisions against a real clock (fast chess) Not measured
  • BulletBench: has not measured this model.

Where to run it

2 hosts on OpenRouter at the standard service tier, with a median of $1.43 blended per million tokens; quantisation where the host states it: FP4, FP8. Cheapest first.

HostQuantisationInputOutputBlended
FeatherlessFP4$1.00$2.70$1.43
StepFunFP8$1.00$2.70$1.43

Prices per million tokens from OpenRouter's public endpoint list, checked 10 Nov 2026.

Against the alternatives

The models you would most likely weigh it against: the leaders of the groups above.

Scroll sideways to see every alternative.

Step 5 PreviewClaude Fable 5.2
leads overall
GPT-6.2 Astra
near the top overall
GPT-6.2 Sol
near the top overall
Claude Opus 5.5
near the top overall
Price per million tokens $1.43 $16.00$16.00$4.00$8.00
Tokens a second – –––96
Intelligence Index – –––57.6
Spring Prompt overall – 88868480
Rank among available models, by use case
Product listings – 8th3rd1st16th
Decks from an analysis – 4th17th6th8th
User surveys – 1st2nd7th5th
Marketing planning – 1st3rd9th11th
Agents and tool use pending –––3rd
Professional work – –––3rd
Reasoning and knowledge – –––1st
Writing 54th of 169 –––5th
Sticking to the facts 54th of 139 –––3rd
Speed under pressure – 10th6th10th–

Shaded figures are better than Step 5 Preview's. A dash means no result.

Where it has been measured

Among models available today, mid-field on writing and sticking to the facts, the use cases measured so far. Results for agents and tool use are pending: those sources last updated before this model was released.

Not timed: Artificial Analysis, our source for speed, has not tested it

Every result

11 published results from 1 source, each in the source's own units, with the configuration that produced it.

Arena (formerly LMArena) · reported by Arena (formerly LMArena) · 11 results
MeasureValueRankConfigurationDated
Arena Agent: confirmed task successIPS effect estimate, higher is better 0.05
0.01–0.08
22 of 50 Step 5 Preview (arena agent) 2 Oct 2026
Arena Agent: praise over complaintIPS effect estimate, higher is better -0.05
-0.10–0.00
38 of 50 Step 5 Preview (arena agent) 2 Oct 2026
Arena Agent: steerabilityIPS effect estimate, higher is better 0.02
-0.00–0.04
21 of 50 Step 5 Preview (arena agent) 2 Oct 2026
Arena Agent: tool groundingIPS effect estimate, higher is better 0.00
0.00–0.01
8 of 50 Step 5 Preview (arena agent) 2 Oct 2026
Arena Text factuality: overallArena rating, higher is better 1,452
1,442–1,461
62 of 182 Step 5 Preview 2 Oct 2026
Arena Text: business, management and financeArena rating, higher is better 1,468
1,442–1,494
42 of 406 Step 5 Preview 2 Oct 2026
Arena Text: creative writingArena rating, higher is better 1,421
1,396–1,445
84 of 411 Step 5 Preview 2 Oct 2026
Arena Text: expert promptsArena rating, higher is better 1,472
1,438–1,505
86 of 363 Step 5 Preview 2 Oct 2026
Arena Text: instruction followingArena rating, higher is better 1,445
1,427–1,464
73 of 413 Step 5 Preview 2 Oct 2026
Arena Text: overallArena rating, higher is better 1,454
1,443–1,465
74 of 413 Step 5 Preview 2 Oct 2026
Arena Text: writing, literature and languageArena rating, higher is better 1,436
1,415–1,457
74 of 412 Step 5 Preview 2 Oct 2026

Not shown: Not a success rate: the estimated effect of choosing this model as the agent's orchestrator, against an even mix of the models tested, inside Arena's own harness.

Test your own prompt.

Benchmarks aren't your task. Run your prompt on the models you could use, with every answer checked and costed.