BenchLeader
Menu

GPT-6 Astra vs Step 5 Preview

Full profiles: GPT-6 Astra (OpenAI) and Step 5 Preview (StepFun): every benchmark, price and rank.

Updated · Data from 13 sources

As of 13 Nov 2026, GPT-6 Astra is ahead of Step 5 Preview on quality (67.1 vs 54.3 on the BenchLeader Index), but Step 5 Preview is 14× cheaper at $1.43 vs $20.00 per 1M tokens blended. Choose GPT-6 Astra for coding (73 vs 50), reasoning (72 vs 52), and knowledge (72 vs 54); choose Step 5 Preview for fast first answers (22.8 s vs 227.9 s) and long context (69 vs 65), where it also streams 3×+ faster.

Key facts (as of 13 Nov 2026)

  • GPT-6 Astra scores 67.1 on the BenchLeader Index and Step 5 Preview scores 54.3 (BenchLeader, 13 Nov 2026).
  • GPT-6 Astra costs $20.00 per 1M tokens blended (list price, 13 Nov 2026).
  • Step 5 Preview costs $1.43 per 1M tokens blended (list price, 13 Nov 2026).
  • Step 5 Preview streams 3×+ faster than GPT-6 Astra (Artificial Analysis, 7-day median, 13 Nov 2026).
  • Of the 58 benchmarks both models have, GPT-6 Astra leads 46 (BenchLeader, 13 Nov 2026).
GPT-6 Astra benchmarks max OpenAI Step 5 Preview benchmarks StepFunAdd a third model

Which to pick for…

The pick's numbers come first. Category scores on the BenchLeader scale (50 is the average model, 15 points one standard deviation). High confidence means a lead of half a standard deviation or more; under one point is called either.

Use casePickWhyConfidence
CodingGPT-6 Astra73 vs 50 category score; FrontierCode 2.0 (AA) 42.4% vs 12.0%High
Agents & tool useGPT-6 Astra64 vs 50 category score; Arena Agent 14.5 vs -3.7High
MathsGPT-6 Astra71 vs 60 category score; Arena Maths 1481 vs 1470High
ReasoningGPT-6 Astra72 vs 52 category score; HLE-Diamond (AA) 63.8% vs 24.3%High
KnowledgeGPT-6 Astra72 vs 54 category score; AA-Omniscience v1.1 42.4 vs 16.4High
Long documentsStep 5 Preview69 vs 65 category score; AA-LCR v1.1 88.3% vs 80.7%Medium
Images and other mediaGPT-6 Astra66 vs 59 category score; Arena Vision 1279 vs 1263Medium
Chat quality (human preference)GPT-6 Astra66 vs 63 on Arena votesMedium
Lowest costStep 5 Preview$1.43 vs $20.00 per 1M blended; agentic coding session $0.071 vs $0.800–
Fastest first answerStep 5 Preview22.8 s (Step 5 Preview) vs 227.9 s (GPT-6 Astra (max)) to the first answer token–
High volume on a budgetNeither flagshipA smaller model is usually the better fit: Cheapest good LLMs, Fastest LLMs–

Key differences

  • GPT-6 Astra (max) leads on quality: 67.1 vs 54.3.
  • GPT-6 Astra (max) is stronger in agents & tools, coding, human preference, knowledge, maths, multimodal, reasoning.
  • Step 5 Preview is stronger in long context.
  • Step 5 Preview is 14× cheaper ($1.43 vs $20.00 per 1M blended).
  • Step 5 Preview streams 3.7× faster (135 vs 36 tokens per second).

At a glance

SpecGPT-6 AstraStep 5 Preview
DeveloperOpenAIStepFun
Released2 Sep 202618 Sep 2026
Context window (tokens)1,050,0001,000,000
Compared atGPT-6 Astra (max)Step 5 Preview
Open weightsNoYes
Price in / out / cached input per 1M$10.00 / $50.00 / $1.00$1.00 / $2.70 / $0.105

Category by category

Category scores on the BenchLeader scale: 50 is the average evaluated model, 15 points is one standard deviation. A gap under one point is called level.

CategoryGPT-6 Astra (max)Step 5 PreviewLeader
Reasoning72.451.7GPT-6 Astra
Coding73.249.9GPT-6 Astra
Agents & tools64.449.7GPT-6 Astra
Maths70.860.1GPT-6 Astra
Knowledge72.353.7GPT-6 Astra
Human preference66.263.1GPT-6 Astra
Multimodal65.659.3GPT-6 Astra
Long context65.069.0Step 5 Preview
Composite–38.3not comparable

On one chart

Each domain on the index's own scale, where 50 is an average model. Click a domain to see the benchmarks behind it. Pick up to 5 models from the list.

2550 avg75ReasoningCodingAgents & toolsMathsKnowledgeLong contextMultimodalHumanpreferenceGPT-6 Astra, Reasoning: 72.4GPT-6 Astra, Coding: 73.2GPT-6 Astra, Agents & tools: 64.4GPT-6 Astra, Maths: 70.8GPT-6 Astra, Knowledge: 72.3GPT-6 Astra, Long context: 65.0GPT-6 Astra, Multimodal: 65.6GPT-6 Astra, Human preference: 66.2Step 5 Preview, Reasoning: 51.7Step 5 Preview, Coding: 49.9Step 5 Preview, Agents & tools: 49.7Step 5 Preview, Maths: 60.1Step 5 Preview, Knowledge: 53.7Step 5 Preview, Long context: 69.0Step 5 Preview, Multimodal: 59.3Step 5 Preview, Human preference: 63.1

Not drawn, measured for fewer than two of these models: Instruction following.

Model, click to add or removeIndex

Where sources disagree

Across the 58 benchmarks both have, GPT-6 Astra leads 46, Step 5 Preview leads 12. Every overall index that rates both puts GPT-6 Astra ahead.

Overall indexGPT-6 Astra (max)Step 5 PreviewLeader
BenchLeader Index67.154.3GPT-6 Astra
AA Intelligence Index v5.056.928.5GPT-6 Astra
Vals Index65.054.8GPT-6 Astra
Arena Text14741449GPT-6 Astra

Each publisher runs its own tasks, harnesses and effort settings, so their overall verdicts can differ; BenchLeader reconciles them on one scale. † measured at another effort setting of the same model (the publisher did not test the one compared here).

Price, speed and latency

List prices blended at three input tokens per output token. Per-task prices use the token counts on the methodology page. Cost, tokens and time per benchmark task are measured: Artificial Analysis’ averages over its whole Intelligence Index run.

MeasureGPT-6 Astra (max)Step 5 Preview
Blended price per 1M tokens$20.00$1.43
Input / output per 1M$10.00 / $50.00$1.00 / $2.70
Cached input per 1M$1.00$0.105
Price per task: chat reply$0.019$0.0012
Price per task: summarize a 30-page report$0.150$0.014
Price per task: code edit$0.135$0.010
Price per task: agentic coding session$0.800$0.071
Price per task: structured extraction$0.030$0.0025
Agentic coding session, input mostly cached$0.395$0.030
Cost per benchmark task (AA Index)$6.24$2.35
Output tokens per task (AA Index)52k113k
Of which reasoning36k80k
Time per task (AA Index)24 min14 min
Output speed36 tok/s135 tok/s
Time to first answer227.9 s22.8 s
Time to a complete answer533.7 s54.4 s

Per token is not per task: an average task of Artificial Analysis' Intelligence Index cost $6.24 on GPT-6 Astra (max) and $2.35 on Step 5 Preview, and Step 5 Preview wrote 2.2× as many output tokens for the same work, which a per-token price hides.

Each model is compared at its best-ranked reasoning-effort setting. GPT-6 Astra (max) scores 67.1 and takes 228 s to its first answer token; GPT-6 Astra scores 66.8 in 5 s, GPT-6 Astra (high) scores 66.3 in 57 s. Step 5 Preview scores 54.3 and takes 23 s to its first answer token. Higher effort means more thinking before the answer: better scores, longer waits and more tokens billed. How reasoning effort changes scores and cost.

Every benchmark

151 benchmarks, grouped by category; each name links to the board and its leaders. † measured at another effort setting of the same model, shown for reference and not counted as a win.

MetricPublisherGPT-6 Astra (max)Step 5 Preview
BenchLeader IndexBenchLeader67.154.3
Agents & tools scoreBenchLeader64.449.7
Coding scoreBenchLeader73.249.9
Human preference scoreBenchLeader66.263.1
Knowledge scoreBenchLeader72.353.7
Long context scoreBenchLeader65.069.0
Maths scoreBenchLeader70.860.1
Multimodal scoreBenchLeader65.659.3
Reasoning scoreBenchLeader72.451.7
Composite scoreBenchLeader–38.3
Blended price $/MBenchLeader$20.00$1.43
Output speedBenchLeader36 tok/s135 tok/s
Time to first answerBenchLeader227.9 s22.8 s
Context windowBenchLeader1,050,000 tokens1,000,000 tokens
Reasoning
GPQA DiamondEpoch AI Benchmarking Hub95.8%–
Humanity's Last ExamScale AI / CAIS54.8%†–
SimpleBenchSimpleBench83.6%†–
Arena Hard PromptsArena (LMArena)15001477
LiveBench 2.0 ReasoningLiveBench92.7%†83.8%
LiveBench 1.0 ReasoningLiveBench92.7%–
GPQA Diamond (AA)Artificial Analysis96.1%–
Humanity's Last Exam (AA)Artificial Analysis54.7%46.5%
HLE-Diamond (AA)Artificial Analysis63.8%24.3%
ARC-AGI-1ARC Prize97.5%–
ARC-AGI-2ARC Prize95.0%–
ARC-AGI-3ARC Prize98.6%–
CritPtArtificial Analysis31.7%20.9%
CritPt v2 hard subset (AA)Artificial Analysis70.5%32.5%
MysteryMechanismVals AI53.1%18.5%
Chess PuzzlesEpoch AI Benchmarking Hub74.0%–
Mystery Game PuzzlesEpoch AI Benchmarking Hub84.0%–
LMCAEpoch AI Benchmarking Hub64.1%†–
DTBenchEpoch AI Benchmarking Hub97.1%†–
ForecastBenchForecastBench66.6%†–
BTF-3Epoch AI Benchmarking Hub13.5%†–
Coding
SciCodeSciCode56.5%58.9%
WeirdMLWeirdML93.3%–
FrontierCodeCognition53.3%–
GSO-BenchGSO-Bench79.4%†–
Show all 151 benchmarks
MetricPublisherGPT-6 Astra (max)Step 5 Preview
Coding
Arena CodingArena (LMArena)15481505
Arena WebDevArena (LMArena)17641573†
LiveBench 2.0 CodingLiveBench87.5%†67.6%
LiveBench 1.0 CodingLiveBench80.4%–
IOIVals AI100.0%38.1%
Code MigrationVals AI67.7%39.0%
Code Migration v1.1Vals AI80.6%47.0%
ProgramBenchVals AI5.5%–
SWE-bench MultimodalVals AI43.3%40.8%
Vibe Code Bench 1-100Vals AI27.6%–
Vibe Code Bench v1.1Vals AI89.6%70.4%
SWE Atlas: Codebase QnAScale AI SEAL59.1%†–
SWE Atlas: RefactoringScale AI SEAL59.0%†–
SWE Atlas: Test WritingScale AI SEAL50.7%†–
MirrorCodeEpoch AI Benchmarking Hub60.0%–
DeepSWE v1.1Epoch AI Benchmarking Hub73.2%–
ALE-BenchEpoch AI Benchmarking Hub2951.3–
FrontierSWEEpoch AI Benchmarking Hub65.5%–
Terminal-Bench 4.0 (AA)Artificial Analysis59.1%33.3%
Terminal-Bench 2.1 (AA)Artificial Analysis88.4%–
Startrise FrontendStartrise79.9%†–
GameBench 2 OverallDuelLab GameBench 288.6†–
GameBench 2 by reasoning settingDuelLab GameBench 294.1†–
Design Arena WebsiteDesign Arena1338–
FrontierCode 1.1 (AA)Artificial Analysis47.0%38.4%
FrontierCode 2.0 (AA)Artificial Analysis42.4%12.0%
Agents & tools
Terminal-BenchTerminal-Bench60.3%–
Remote Labor IndexScale AI / CAIS20.8%†–
APEX-AgentsMercor64.7%†–
Arena AgentArena (LMArena)14.5-3.7
LiveBench Long AgenticLiveBench75.6%†19.4%
LiveBench BrowserLiveBench91.3%†24.8%
LiveBench Agentic CodingLiveBench57.3%–
Terminal-Bench 2.1 (Vals)Vals AI87.3%–
HiL-BenchScale AI SEAL65.3%†–
GDPval-AA v2.1, normalised scoreArtificial Analysis53.7%54.3%
GDPval-AA v2.1Artificial Analysis15861592
τ³-Banking (AA)Artificial Analysis41.4%–
ITBench SRE (AA)Artificial Analysis48.6%55.6%
Analyst Agent (AA)Artificial Analysis51.3%35.0%
APEX-Agents (AA)Artificial Analysis–38.0%
CUA-benchVals AI19.2%–
CyberBenchVals AI41.1%–
Finance Agent v2Vals AI53.5%50.7%
Finance Agent v2.1Vals AI60.4%58.9%
Harvey's Legal Agent Benchmark v1.1Vals AI7.1%12.9%
Harvey's Legal Agent Benchmark v1Vals AI5.4%9.2%
Legal Research BenchVals AI39.4%40.4%
Robo-benchVals AI23.3%–
SREBenchVals AI56.9%0.4%
Tax Agent BenchVals AI20.7%27.3%
Terminal-Bench 4.0 (Vals)Vals AI59.6%31.8%
Terminal-Bench ScienceVals AI62.9%2.9%
Time Horizon Index: KSPVals AI90.5%–
Arena SearchArena (LMArena)1245–
BALROGBALROG68.3%–
PostTrainBenchEpoch AI Benchmarking Hub41.9%–
Vending-Bench 2Andon Labs15514.7†–
AutomationBenchArtificial Analysis68.5%51.0%
GDP.pdf (AA)Artificial Analysis31.0%14.8%
Harvey LAB-AA v1.1Harvey8.6%–
EnterpriseOps-GymArtificial Analysis–47.2%
AA-Briefcase v1.1Artificial Analysis15691423
GDPxGDPx by SemiAnalysis9.6%–
Design Arena AgenticDesign Arena1393–
RoboDojo (LLM as robot policy)RoboDojo29.0†–
Terminal-Bench-Science 0.1 (AA)Artificial Analysis63.3%2.4%
CWE-Bench (AA)Artificial Analysis63.3%52.5%
DeepSec-Bench (AA)Artificial Analysis36.9%15.8%
CyberGym end-to-end (AA)Artificial Analysis0.0%74.8%
FACTS SearchGoogle93.3%†–
Maths
FrontierMath Tiers 1–3Epoch AI Benchmarking Hub93.7%–
FrontierMath Tier 4Epoch AI Benchmarking Hub97.6%–
OTIS Mock AIMEEpoch AI Benchmarking Hub100.0%–
ProofBenchVals AI99.0%†42.0%
LiveBench MathematicsLiveBench96.8%–
Arena MathsArena (LMArena)14811470
Knowledge
SimpleQA VerifiedEpoch AI Benchmarking Hub75.6%–
LiveBench 2.0 Data AnalysisLiveBench83.0%†63.5%
LiveBench ResearchLiveBench90.6%†75.5%
LiveBench 1.0 Data AnalysisLiveBench83.0%–
AA-Omniscience v1.1Artificial Analysis42.416.4
PRBench FinanceScale AI SEAL47.5%†–
PRBench LegalScale AI SEAL48.4%†–
BioMysteryBenchVals AI79.3%67.4%
Excel Modeling BenchmarkVals AI71.7%57.1%
Excel Modeling Benchmark v1.1Vals AI72.7%69.5%
MedCodeVals AI48.5%–
MedScribeVals AI87.9%–
Public Benefits Bench v1.1Vals AI–58.1%
DrugDiscoveryBenchScale AI SEAL68.7%–
EBR-benchEpoch AI Benchmarking Hub76.2%–
GDP.pdfEpoch AI Benchmarking Hub34.2%–
AA-Omniscience v1.1: accuracyArtificial Analysis61.6%36.4%
AA-Omniscience v1.1: non-hallucinationArtificial Analysis50.2%68.7%
HealthBench Professional (AA)OpenAI66.0%–
Instruction following
LiveBench LanguageLiveBench89.4%–
LiveBench Instruction FollowingLiveBench75.6%–
Arena Instruction FollowingArena (LMArena)14741450
Human preference
Arena TextArena (LMArena)14741449
Arena Creative WritingArena (LMArena)14511413
Arena Multi-turnArena (LMArena)14831445
Arena Longer QueriesArena (LMArena)14851461
Multimodal
Arena VisionArena (LMArena)12791263
LiveBench DesignLiveBench80.0%†70.8%
MMMU-ProArtificial Analysis86.9%76.4%
Blueprint-Bench 2Epoch AI Benchmarking Hub49.7%†–
Code2Video-AAArtificial Analysis1427821
AA-3DArtificial Analysis1168859
Long context
AA-LCR v1.1Artificial Analysis80.7%88.3%
Arena DocumentArena (LMArena)1474–
MLCRArtificial Analysis41.7%18.3%
Composite
Epoch Capabilities IndexEpoch AI Benchmarking Hub169†–
LiveBench 2.0LiveBench84.5%†54.9%
LiveBench 1.0LiveBench82.2%–
AA Intelligence Index v5.0Artificial Analysis56.928.5
AA Intelligence Index v4.3.2Artificial Analysis52.743.7
AA Progress Index v1Artificial Analysis349.6318.2
Vals IndexVals AI65.054.8
Vals Finance IndexVals AI55.6%55.4%
Vals RSI IndexVals AI27.1%–
Vals Science IndexVals AI65.1%29.6%
Vals Router IndexVals AI65.0%†–
AA Cyber IndexArtificial Analysis33.447.7
Safety & honesty
FORTRESSScale AI SEAL12.1%†–
Metadata
AA Openness IndexArtificial Analysis–33.3

Data as of 2026-11-13. Best configuration of each model; every score links to its source on the model pages.

DuelLab GameBench 2, release gb2-20261110-2acac640a228 (https://benchmarks.duellab.org/releases/gb2-20261110-2acac640a228/), CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Scores mapped to BenchLeader model names; not endorsed by DuelLab.

What changed

  • GPT-6 Astra (max): 1 new result (PostTrainBench); 2 score revisions.
  • Step 5 Preview: 1 score revision.
  • GPT-6 Astra (max): 1 new result (MirrorCode); 1 score revision.
  • GPT-6 Astra (max): BenchLeader Index 68.1 → 67.4.
  • GPT-6 Astra (max): 2 score revisions.
  • Step 5 Preview: 7 score revisions.
  • GPT-6 Astra (max): 1 score revision.
  • GPT-6 Astra (max): 1 score revision.
  • GPT-6 Astra (max): 1 score revision.
  • GPT-6 Astra (max): 1 new result (MirrorCode).

Rankings: Best LLM overall · Best LLM for coding · Best LLM for agents · Best cheap LLMs

GPT-6 Astra vs Step 5 Preview: questions

Is GPT-6 Astra better than Step 5 Preview?
GPT-6 Astra (max) leads on quality: 67.1 vs 54.3. The BenchLeader Index combines every independent quality benchmark; GPT-6 Astra (max) is ahead overall as of 13 Nov 2026, but check the category scores for your use.
GPT-6 Astra vs Step 5 Preview: which is cheaper, and will that change?
Step 5 Preview is cheaper today: $1.43 against $20.00 per million tokens, blended at three input tokens per output token. No price change has been announced for either.
Is GPT-6 Astra or Step 5 Preview better for coding?
GPT-6 Astra scores higher in coding (73 vs 50 on the category index, where 50 is average). On FrontierCode 2.0 (AA), the top shared coding benchmark, it is 42.4% for GPT-6 Astra against 12.0% for Step 5 Preview.
Which is faster, GPT-6 Astra or Step 5 Preview?
Step 5 Preview streams 3×+ faster (135 vs 36 output tokens per second, Artificial Analysis 7-day median).
Is GPT-6 Astra or Step 5 Preview better for agents and tool use?
GPT-6 Astra scores higher in agentic tasks (64 vs 50 on the category index, where 50 is average). On Arena Agent, the top shared agents & tools benchmark, it is 14.5 for GPT-6 Astra against -3.7 for Step 5 Preview.
More questions (4)
Which is better for long documents?
Step 5 Preview scores higher on long-context benchmarks (69 vs 65). Both take about 1.1M tokens: GPT-6 Astra 1,050,000, Step 5 Preview 1,000,000.
Which has lower latency?
Step 5 Preview answers first: 22.8 s against 227.9 s to the first answer token, as Artificial Analysis measures it. Reasoning effort drives most of this: a higher setting thinks longer before it answers.
Do independent benchmarks agree on GPT-6 Astra vs Step 5 Preview?
Across the 58 benchmarks both have, GPT-6 Astra leads 46, Step 5 Preview leads 12. Every overall index that rates both puts GPT-6 Astra ahead.
When was Step 5 Preview released?
Step 5 Preview was released on 18 Sep 2026.