BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases16
StepFunOpen weightsReasoning modelCompare with another model

Step 5 Preview

Reasoning effort

Step 5 Preview is a StepFun open-weights reasoning model, released 18 Sept 2026. Its best configuration ranks #108 of 775 on the BenchLeader Index at 60.6 ±5.4, in the upper half. The ± is the point: 220 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in reasoning (72) and lowest in agents & tools (51). At $1.43 per million tokens blended it is pricier than most ranked models. Output speed of 92 tokens per second puts it faster than most, with a first answer in 24.1 s. It has been measured at 2 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 19 Oct 2026.

Blended price
$1.43/M
$1.00 in · $2.70 out
Output speed
92 tok/s
measured by Artificial Analysis
First answer
24 s
first token 1.19 s
Context
1M
Full answer
29 s
median, reasoning included
Cost per run
$1.03
one full Intelligence Index run
Released
18 Sept 2026
How it scores by categoryDashed line = average model (50). One step of 15 = one standard deviation.
  1. Overall index61
  2. Reasoning72
  3. Coding56
  4. Agents & tools51
  5. Maths60
  6. Knowledge69
  7. Human preference64
  8. Multimodal60
  9. Long context69
  10. Composite61

Reasoning-effort configurations

The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.

EffortIndexRankSpeedFirst answerChat reply costCategories
high––92 tok/s24 s$0.0012Coding 60
not statedbest60.6#10892 tok/s24 s$0.0012Agents & tools 51 · Coding 56 · Composite 61 · Human preference 64 · Knowledge 69 · Long context 69 · Maths 60 · Multimodal 60 · Reasoning 72

How its index has moved

19 Sept 2026 to 19 Oct 2026
556371

The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.

Benchmark results

See all LLM benchmarks. One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.

Metadata

Benchmarknot statedSourceTrend
AA Openness Indexnot in index33.3#32Artificial Analysis

Reasoning

Benchmarknot statedSourceTrend
Arena Hard Prompts1475#73Arena (LMArena)
LiveBench 2.0 Reasoningnot in index83.8%#33LiveBench
Humanity's Last Exam (AA)not in index46.5%#43Artificial Analysis
CritPt20.9%#56Artificial Analysis
MysteryMechanismnot in index18.5%#24Vals AI

Coding

Benchmarkhighnot statedSourceTrend
SciCodeeffort not stated–58.9%#16SciCode
Arena Codingeffort not stated–1508#63Arena (LMArena)
Arena WebDev1567#36–Arena (LMArena)
LiveBench 2.0 Codingnot in indexeffort not stated–85.6%#21LiveBench
SciCode (AA)not in indexeffort not stated–58.9%#17Artificial Analysis
IOIeffort not stated–38.1%#37Vals AI
Code Migrationnot in indexeffort not stated–39.0%#31Vals AI
Vibe Code Bench v1.1not in indexeffort not stated–70.4%#40Vals AI
Terminal-Bench 4.0 (AA)not in indexeffort not stated–33.3%#38Artificial Analysis
FrontierCode 1.1 (AA)not in indexeffort not stated–38.4%#30Artificial Analysis

Agents & tools

Benchmarknot statedSourceTrend
Arena Agent-1.8#33Arena (LMArena)
LiveBench Long Agenticnot in index27.5%#36LiveBench
LiveBench Browsernot in index24.0%#38LiveBench
GDPval-AA v2.1not in index54.3%#37Artificial Analysis
ITBench SRE (AA)not in index55.6%#2Artificial Analysis
Analyst Agent (AA)not in index35.0%#28Artificial Analysis
APEX-Agents (AA)not in index38.0%#5Artificial Analysis
Finance Agent v2not in index50.7%#41Vals AI
Legal Research Benchnot in index40.4%#29Vals AI
SREBenchnot in index0.4%#29Vals AI
Tax Agent Benchnot in index27.3%#32Vals AI
Terminal-Bench 4.0 (Vals)not in index31.8%#17Vals AI
Terminal-Bench Sciencenot in index2.9%#32Vals AI
AutomationBenchnot in index51.0%#75Artificial Analysis
GDP.pdf (AA)not in index14.8%#99Artificial Analysis
EnterpriseOps-Gymnot in index47.2%#10Artificial Analysis
AA-Briefcase v1.1not in index1424#39Artificial Analysis
Terminal-Bench-Science 0.1 (AA)not in index2.4%#44Artificial Analysis
CWE-Bench (AA)52.5%#11Artificial Analysis
DeepSec-Bench (AA)15.8%#13Artificial Analysis
CyberGym end-to-end (AA)74.8%#4Artificial Analysis

Maths

Benchmarknot statedSourceTrend
ProofBench42.0%#37Vals AI
Arena Maths1459#63Arena (LMArena)

Knowledge

Benchmarknot statedSourceTrend
LiveBench 2.0 Data Analysisnot in index63.5%#40LiveBench
LiveBench Researchnot in index72.8%#25LiveBench
AA-Omniscience16.4#78Artificial Analysis
BioMysteryBenchnot in index67.4%#16Vals AI
Excel Modeling Benchmarknot in index57.1%#40Vals AI
Public Benefits Benchnot in index58.1%#35Vals AI
AA-Omniscience: accuracynot in index41.5%#109Artificial Analysis
AA-Omniscience: non-hallucinationnot in index57.0%#86Artificial Analysis

Instruction following

Benchmarknot statedSourceTrend
Arena Instruction Followingnot in index1441#78Arena (LMArena)

Human preference

Benchmarknot statedSourceTrend
Arena Text1447#82Arena (LMArena)
Arena Creative Writingnot in index1403#102Arena (LMArena)
Arena Multi-turnnot in index1447#92Arena (LMArena)
Arena Longer Queriesnot in index1455#79Arena (LMArena)

Multimodal

Benchmarknot statedSourceTrend
Arena Vision1265#49Arena (LMArena)
LiveBench Designnot in index76.9%#24LiveBench
MMMU-Pro76.4%#88Artificial Analysis
AA-3Dnot in index852#18Artificial Analysis

Long context

Benchmarknot statedSourceTrend
AA-LCR88.3%#2Artificial Analysis
MLCRnot in index16.7%#40Artificial Analysis

Composite

Benchmarknot statedSourceTrend
LiveBench 2.062.0%#37LiveBench
AA Intelligence Index v4.3.243.7#44Artificial Analysis
AA Cyber Indexnot in index47.7#6Artificial Analysis

Where it wins

Benchmarks where this configuration ranks in the top five of every configuration measured.

Where to run it

Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.

ProviderSpeedFirst tokenInput $/MOutput $/MContextQuantisation
inference.net198 tok/s0.53 s$1.00$2.701.0Mfp4
inference.net143 tok/s0.80 s$1.00$2.701.0Mfp4
StepFun105 tok/s1.35 s$1.00$2.701Mfp8
DeepInfra72 tok/s1.82 s$1.00$2.701.0Mfp8
Featherless38 tok/s2.96 s$1.00$2.70262kfp4

What a task costs

Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache at $0.048 per 1M. Reasoning tokens are not modelled.

WorkloadTokens in / outCostWith cachingTime
Chat reply400 / 300$0.0012$0.000927.3 s
Summarize a 30-page report12,000 / 600$0.014$0.005130.6 s
Code edit6,000 / 1,500$0.010$0.005840.3 s
Agentic coding session60,000 / 4,000$0.071$0.0281.1 min
Structured extraction2,000 / 200$0.0025$0.001126.2 s

More from StepFun

New this month

Compare Step 5 Preview with…

See also

Best under $2/MBest open weightsCheapest 1M contextBest for codingAll StepFun models

Data as of 21 Oct 2026. Compare these configurations.

Cite as: BenchLeader, “Step 5 Preview: benchmarks, pricing, speed and rank”, https://www.benchleader.com/models/step-5, data as of 21 Oct 2026.