Claude Opus 4.8 vs Step 5 Preview
Full profiles: Claude Opus 4.8 (Anthropic) and Step 5 Preview (StepFun): every benchmark, price and rank.
Updated · Data from 9 sources
As of 13 Nov 2026, Claude Opus 4.8 and Step 5 Preview are neck and neck on quality (60.4 vs 54.3 on the BenchLeader Index, inside the ±7.2 margin), and Step 5 Preview is 7.0× cheaper at $1.43 vs $10.00 per 1M tokens blended. Claude Opus 4.8 leads in every category we measure both in. Step 5 Preview streams ~2× faster.
Key facts (as of 13 Nov 2026)
- Claude Opus 4.8 scores 60.4 on the BenchLeader Index and Step 5 Preview scores 54.3 (BenchLeader, 13 Nov 2026).
- Claude Opus 4.8 costs $10.00 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 ~2× faster than Claude Opus 4.8 (Artificial Analysis, 7-day median, 13 Nov 2026).
- Of the 8 benchmarks both models have, Claude Opus 4.8 leads 8 (BenchLeader, 13 Nov 2026).
Claude Opus 4.8 benchmarks 
Step 5 Preview benchmarks 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 case | Pick | Why | Confidence |
|---|---|---|---|
| Coding | Claude Opus 4.8 | 61 vs 50 category score; Arena Coding 1534 vs 1505 | High |
| Maths | Claude Opus 4.8 | 68 vs 60 category score; Arena Maths 1492 vs 1470 | Medium |
| Reasoning | Claude Opus 4.8 | 60 vs 52 category score; Arena Hard Prompts 1513 vs 1477 | High |
| Images and other media | Claude Opus 4.8 | 64 vs 59 category score; Arena Vision 1295 vs 1263 | Medium |
| Chat quality (human preference) | Claude Opus 4.8 | 67 vs 63 on Arena votes | Medium |
| Lowest cost | Step 5 Preview | $1.43 vs $10.00 per 1M blended; agentic coding session $0.071 vs $0.400 | – |
| Fastest first answer | Claude Opus 4.8 | 0.8 s (Claude Opus 4.8 (high)) vs 22.8 s (Step 5 Preview) to the first answer token | – |
| High volume on a budget | Neither flagship | A smaller model is usually the better fit: Cheapest good LLMs, Fastest LLMs | – |
Key differences
- Claude Opus 4.8 (high) and Step 5 Preview are neck and neck on quality (60.4 vs 54.3, within the ±7.2 margin).
- Claude Opus 4.8 (high) is stronger in coding, human preference, maths, multimodal, reasoning.
- Step 5 Preview is 7.0× cheaper ($1.43 vs $10.00 per 1M blended).
- Step 5 Preview streams 2.0× faster (135 vs 68 tokens per second).
At a glance
| Spec | Claude Opus 4.8 | Step 5 Preview |
|---|---|---|
| Developer | Anthropic | StepFun |
| Released | 28 May 2026 | 18 Sep 2026 |
| Context window (tokens) | 1,000,000 | 1,000,000 |
| Compared at | Claude Opus 4.8 (high) | Step 5 Preview |
| Open weights | No | Yes |
| Price in / out / cached input per 1M | $5.00 / $25.00 / $0.500 | $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.
| Category | Claude Opus 4.8 (high) | Step 5 Preview | Leader |
|---|---|---|---|
| Reasoning | 59.8 | 51.7 | Claude Opus 4.8 |
| Coding | 60.8 | 49.9 | Claude Opus 4.8 |
| Agents & tools | – | 49.7 | not comparable |
| Maths | 67.5 | 60.1 | Claude Opus 4.8 |
| Knowledge | – | 53.7 | not comparable |
| Human preference | 67.2 | 63.1 | Claude Opus 4.8 |
| Multimodal | 64.0 | 59.3 | Claude Opus 4.8 |
| Long context | – | 69.0 | not comparable |
| Composite | – | 38.3 | not 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.
Not drawn, measured for fewer than two of these models: Agents & tools, Knowledge, Instruction following, Long context.
Model, click to add or removeIndex
Where sources disagree
Across the 8 benchmarks both have, Claude Opus 4.8 leads 8, Step 5 Preview leads 0. Every overall index that rates both puts Claude Opus 4.8 ahead.
| Overall index | Claude Opus 4.8 (high) | Step 5 Preview | Leader |
|---|---|---|---|
| BenchLeader Index | 60.4 | 54.3 | neck and neck |
| Arena Text | 1482 | 1449 | Claude Opus 4.8 |
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.
| Measure | Claude Opus 4.8 (high) | Step 5 Preview |
|---|---|---|
| Blended price per 1M tokens | $10.00 | $1.43 |
| Input / output per 1M | $5.00 / $25.00 | $1.00 / $2.70 |
| Cached input per 1M | $0.500 | $0.105 |
| Price per task: chat reply | $0.0095 | $0.0012 |
| Price per task: summarize a 30-page report | $0.075 | $0.014 |
| Price per task: code edit | $0.068 | $0.010 |
| Price per task: agentic coding session | $0.400 | $0.071 |
| Price per task: structured extraction | $0.015 | $0.0025 |
| Agentic coding session, input mostly cached | $0.198 | $0.030 |
| Cost per benchmark task (AA Index) | – | $2.35 |
| Output tokens per task (AA Index) | – | 113k |
| Of which reasoning | – | 80k |
| Time per task (AA Index) | – | 14 min |
| Output speed | 68 tok/s | 135 tok/s |
| Time to first answer | 0.8 s | 22.8 s |
| Time to a complete answer | – | 54.4 s |
Each model is compared at its best-ranked reasoning-effort setting. Claude Opus 4.8 (high) scores 60.4 and takes 1 s to its first answer token; Claude Opus 4.8 (max) scores 60.0 in 36 s, Claude Opus 4.8 scores 59.5 in 1 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
164 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.
| Metric | Publisher | Claude Opus 4.8 (high) | Step 5 Preview |
|---|---|---|---|
| BenchLeader Index | BenchLeader | 60.4 | 54.3 |
| Coding score | BenchLeader | 60.8 | 49.9 |
| Human preference score | BenchLeader | 67.2 | 63.1 |
| Maths score | BenchLeader | 67.5 | 60.1 |
| Multimodal score | BenchLeader | 64.0 | 59.3 |
| Reasoning score | BenchLeader | 59.8 | 51.7 |
| Agents & tools score | BenchLeader | – | 49.7 |
| Composite score | BenchLeader | – | 38.3 |
| Knowledge score | BenchLeader | – | 53.7 |
| Long context score | BenchLeader | – | 69.0 |
| Blended price $/M | BenchLeader | $10.00 | $1.43 |
| Output speed | BenchLeader | 68 tok/s | 135 tok/s |
| Time to first answer | BenchLeader | 0.8 s | 22.8 s |
| Context window | BenchLeader | 1,000,000 tokens | 1,000,000 tokens |
| Reasoning | |||
| GPQA Diamond | Epoch AI Benchmarking Hub | 91.0%† | – |
| SimpleBench | SimpleBench | 64.8%† | – |
| Arena Hard Prompts | Arena (LMArena) | 1513 | 1477 |
| LiveBench 2.0 Reasoning | LiveBench | – | 83.8% |
| LiveBench 1.0 Reasoning | LiveBench | 89.2%† | – |
| GPQA Diamond (AA) | Artificial Analysis | 92.0%† | – |
| Humanity's Last Exam (AA) | Artificial Analysis | 48.7%† | 46.5% |
| HLE-Diamond (AA) | Artificial Analysis | – | 24.3% |
| GPQA Diamond (Vals) | Vals AI | 92.4%† | – |
| EnigmaEval | Scale AI SEAL | 23.5%† | – |
| Kagi LLM Benchmark | Kagi LLM Benchmark | 88.8%† | – |
| ARC-AGI-1 | ARC Prize | 92.0% | – |
| ARC-AGI-2 | ARC Prize | 72.1% | – |
| ARC-AGI-3 | ARC Prize | 1.5% | – |
| CritPt | Artificial Analysis | 20.9%† | 20.9% |
| CritPt v2 hard subset (AA) | Artificial Analysis | – | 32.5% |
| MysteryMechanism | Vals AI | – | 18.5% |
| Chess Puzzles | Epoch AI Benchmarking Hub | 52.0%† | – |
| Mystery Game Puzzles | Epoch AI Benchmarking Hub | 36.0%† | – |
| LMCA | Epoch AI Benchmarking Hub | 57.5%† | – |
| DTBench | Epoch AI Benchmarking Hub | 94.9%† | – |
| ForecastBench | ForecastBench | 62.4%† | – |
| BTF-3 | Epoch AI Benchmarking Hub | 14.0% | – |
| Coding | |||
| SciCode | SciCode | 53.5%† | 58.9% |
| WeirdML | WeirdML | 70.5%† | – |
Show all 164 benchmarks
| Metric | Publisher | Claude Opus 4.8 (high) | Step 5 Preview |
|---|---|---|---|
| Coding | |||
| FrontierCode | Cognition | 46.5%† | – |
| GSO-Bench | GSO-Bench | 47.1%† | – |
| Arena Coding | Arena (LMArena) | 1534 | 1505 |
| Arena WebDev | Arena (LMArena) | 1555 | 1573† |
| LiveBench 2.0 Coding | LiveBench | – | 67.6% |
| LiveBench 1.0 Coding | LiveBench | 81.8%† | – |
| SciCode (AA) | Artificial Analysis | 54.4%† | 58.9% |
| LiveCodeBench | Vals AI | 87.8%† | – |
| IOI | Vals AI | – | 38.1% |
| SWE-bench (Vals) | Vals AI | 88.6%† | – |
| Code Migration | Vals AI | 47.3%† | 39.0% |
| Code Migration v1.1 | Vals AI | – | 47.0% |
| ProgramBench | Vals AI | 1.0%† | – |
| SWE-bench Multimodal | Vals AI | – | 40.8% |
| Vibe Code Bench v1.1 | Vals AI | 82.7%† | 70.4% |
| SWE Atlas: Codebase QnA | Scale AI SEAL | 57.3%† | – |
| SWE Atlas: Refactoring | Scale AI SEAL | 46.7%† | – |
| SWE Atlas: Test Writing | Scale AI SEAL | 49.6%† | – |
| MirrorCode | Epoch AI Benchmarking Hub | 65.6%† | – |
| Surface Evolver Bench | Epoch AI Benchmarking Hub | 87.5% | – |
| DeepSWE v1.1 | Epoch AI Benchmarking Hub | 51.8% | – |
| ALE-Bench | Epoch AI Benchmarking Hub | 1563.8 | – |
| Terminal-Bench 4.0 (AA) | Artificial Analysis | 21.7%† | 33.3% |
| Terminal-Bench 2.1 (AA) | Artificial Analysis | 84.6%† | – |
| GameBench 2 Overall | DuelLab GameBench 2 | 49.6† | – |
| GameBench 2 by reasoning setting | DuelLab GameBench 2 | 46.0† | – |
| Design Arena Website | Design Arena | 1250† | – |
| FrontierCode 1.1 (AA) | Artificial Analysis | – | 38.4% |
| FrontierCode 2.0 (AA) | Artificial Analysis | – | 12.0% |
| Agents & tools | |||
| Terminal-Bench | Terminal-Bench | 23.6%† | – |
| OSWorld-Verified 2.0 | OSWorld | 20.6%† | – |
| Remote Labor Index | Scale AI / CAIS | 8.3%† | – |
| APEX-Agents | Mercor | 48.9%† | – |
| Arena Agent | Arena (LMArena) | – | -3.7 |
| LiveBench Long Agentic | LiveBench | – | 19.4% |
| LiveBench Browser | LiveBench | – | 24.8% |
| LiveBench Agentic Coding | LiveBench | 50.5%† | – |
| Terminal-Bench Hard | Artificial Analysis | 58.3%† | – |
| τ²-Bench Telecom (AA) | Artificial Analysis | 94.4%† | – |
| Terminal-Bench 2.1 (Vals) | Vals AI | 71.9%† | – |
| MCP Atlas | Scale AI SEAL | 82.2%† | – |
| HiL-Bench | Scale AI SEAL | 35.3%† | – |
| τ³-bench Banking | τ²-bench | 39.7%† | – |
| GDPval-AA v2.1, normalised score | Artificial Analysis | – | 54.3% |
| GDPval-AA v2.1 | Artificial Analysis | – | 1592 |
| τ³-Banking (AA) | Artificial Analysis | 34.2%† | – |
| ITBench SRE (AA) | Artificial Analysis | – | 55.6% |
| Analyst Agent (AA) | Artificial Analysis | 45.0%† | 35.0% |
| APEX-Agents (AA) | Artificial Analysis | – | 38.0% |
| Finance Agent v2 | Vals AI | 53.9%† | 50.7% |
| Finance Agent v2.1 | Vals AI | – | 58.9% |
| Harvey's Legal Agent Benchmark v1.1 | Vals AI | – | 12.9% |
| Harvey's Legal Agent Benchmark v1 | Vals AI | 9.6%† | 9.2% |
| Legal Research Bench | Vals AI | 43.8%† | 40.4% |
| SkillsBench | Vals AI | 59.2%† | – |
| SREBench | Vals AI | – | 0.4% |
| Tax Agent Bench | Vals AI | 30.0%† | 27.3% |
| Terminal-Bench 2.0 (Vals) | Vals AI | 70.0%† | – |
| Terminal-Bench 4.0 (Vals) | Vals AI | 23.2%† | 31.8% |
| Terminal-Bench Science | Vals AI | 4.3%† | 2.9% |
| Time Horizon Index: KSP | Vals AI | 11.8%† | – |
| Arena Search | Arena (LMArena) | 1194† | – |
| DeepResearch Bench | DeepResearch Bench | 50.2% | – |
| PostTrainBench | Epoch AI Benchmarking Hub | 30.7% | – |
| GBAEval | Epoch AI Benchmarking Hub | 70.9%† | – |
| Vending-Bench 2 | Andon Labs | 5787.4† | – |
| AutomationBench | Artificial Analysis | 45.6%† | 51.0% |
| GDP.pdf (AA) | Artificial Analysis | 22.8%† | 14.8% |
| EnterpriseOps-Gym | Artificial Analysis | 44.0%† | 47.2% |
| AA-Briefcase v1.1 | Artificial Analysis | 1320† | 1423 |
| Design Arena Agentic | Design Arena | 1218† | – |
| Terminal-Bench-Science 0.1 (AA) | Artificial Analysis | – | 2.4% |
| CWE-Bench (AA) | Artificial Analysis | – | 52.5% |
| DeepSec-Bench (AA) | Artificial Analysis | – | 15.8% |
| CyberGym end-to-end (AA) | Artificial Analysis | – | 74.8% |
| Maths | |||
| FrontierMath Tiers 1–3 | Epoch AI Benchmarking Hub | 80.0%† | – |
| FrontierMath Tier 4 | Epoch AI Benchmarking Hub | 56.1%† | – |
| OTIS Mock AIME | Epoch AI Benchmarking Hub | 98.3%† | – |
| ProofBench | Vals AI | 69.0%† | 42.0% |
| LiveBench Mathematics | LiveBench | 94.3%† | – |
| AIME 2026 | MathArena | 100.0%† | – |
| HMMT February 2026 | MathArena | 95.5%† | – |
| MathArena Apex | MathArena | 81.3%† | – |
| Arena Maths | Arena (LMArena) | 1492 | 1470 |
| Knowledge | |||
| SimpleQA Verified | Epoch AI Benchmarking Hub | 53.0%† | – |
| LiveBench 2.0 Data Analysis | LiveBench | – | 63.5% |
| LiveBench Research | LiveBench | – | 75.5% |
| LiveBench 1.0 Data Analysis | LiveBench | 66.0%† | – |
| AA-Omniscience v1.1 | Artificial Analysis | – | 16.4 |
| MMLU-Pro | Vals AI | 89.6%† | – |
| LegalBench | Vals AI | 83.6%† | – |
| CorpFin | Vals AI | 66.7%† | – |
| TaxEval | Vals AI | 75.6%† | – |
| BioMysteryBench | Vals AI | – | 67.4% |
| Excel Modeling Benchmark | Vals AI | 69.4%† | 57.1% |
| Excel Modeling Benchmark v1.1 | Vals AI | – | 69.5% |
| MedCode | Vals AI | 53.2%† | – |
| MedScribe | Vals AI | 85.8%† | – |
| Public Benefits Bench v1.1 | Vals AI | 68.1%† | 58.1% |
| DrugDiscoveryBench | Scale AI SEAL | 46.8%† | – |
| EBR-bench | Epoch AI Benchmarking Hub | 28.6%† | – |
| GDP.pdf | Epoch AI Benchmarking Hub | 24.0%† | – |
| AA-Omniscience v1.1: accuracy | Artificial Analysis | – | 36.4% |
| AA-Omniscience v1.1: non-hallucination | Artificial Analysis | – | 68.7% |
| Instruction following | |||
| LiveBench Language | LiveBench | 79.7%† | – |
| LiveBench Instruction Following | LiveBench | 72.0%† | – |
| IFBench | Artificial Analysis | 62.2%† | – |
| Arena Instruction Following | Arena (LMArena) | 1492 | 1450 |
| Human preference | |||
| Arena Text | Arena (LMArena) | 1482 | 1449 |
| EQ-Bench 4 | EQ-Bench | 1268† | – |
| Arena Creative Writing | Arena (LMArena) | 1470 | 1413 |
| Arena Multi-turn | Arena (LMArena) | 1497 | 1445 |
| Arena Longer Queries | Arena (LMArena) | 1503 | 1461 |
| Multimodal | |||
| Arena Vision | Arena (LMArena) | 1295 | 1263 |
| LiveBench Design | LiveBench | – | 70.8% |
| MMMU-Pro | Artificial Analysis | – | 76.4% |
| MMMU-Pro (Vals) | Vals AI | 86.6%† | – |
| MortgageTax | Vals AI | 69.9%† | – |
| GeoBench | GeoBench | 81.0% | – |
| Blueprint-Bench 2 | Epoch AI Benchmarking Hub | 14.5%† | – |
| Code2Video-AA | Artificial Analysis | – | 821 |
| AA-3D | Artificial Analysis | – | 859 |
| Long context | |||
| AA-LCR v1.1 | Artificial Analysis | 77.7%† | 88.3% |
| Arena Document | Arena (LMArena) | 1485 | – |
| MLCR | Artificial Analysis | 19.4%† | 18.3% |
| Composite | |||
| Epoch Capabilities Index | Epoch AI Benchmarking Hub | 159.9† | – |
| LiveBench 2.0 | LiveBench | – | 54.9% |
| LiveBench 1.0 | LiveBench | 76.2%† | – |
| AA Intelligence Index v5.0 | Artificial Analysis | – | 28.5 |
| AA Intelligence Index v4.3.2 | Artificial Analysis | 41.8† | 43.7 |
| AA Progress Index v1 | Artificial Analysis | 318.4† | 318.2 |
| Vals Index | Vals AI | – | 54.8 |
| Vals Finance Index | Vals AI | – | 55.4% |
| Vals RSI Index | Vals AI | 24.0%† | – |
| Vals Science Index | Vals AI | – | 29.6% |
| Vals Multimodal Index | Vals AI | 70.9%† | – |
| AA Cyber Index | Artificial Analysis | – | 47.7 |
| Safety & honesty | |||
| FORTRESS | Scale AI SEAL | 18.2%† | – |
| Metadata | |||
| AA Openness Index | Artificial 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
- Claude Opus 4.8 (high): 1 score revision.
- Step 5 Preview: 1 score revision.
- Claude Opus 4.8 (high): 1 new result (GeoBench).
- Step 5 Preview: 7 score revisions.
- Step 5 Preview: 2 new results (Vals Index, Harvey's Legal Agent Benchmark v1.1).
- Step 5 Preview: 1 score revision.
- Step 5 Preview: BenchLeader Index 58.0 → 54.7.
- Step 5 Preview: 5 new results (AA Intelligence Index v5.0, AA Progress Index v1, HLE-Diamond (AA)…); 2 score revisions; Step 5 Preview cost per task moved from $1.03/M to $2.35/M (128% up).
- Step 5 Preview: 3 score revisions.
- Step 5 Preview: 1 score revision.
Related comparisons
- Claude Opus 4.8 vs Gemini 4 Carbon
- Step 5 Preview vs Claude Opus 5.5
- Step 5 Preview vs GPT-6 Astra
- Step 5 Preview vs GPT-6.1 Sol
- Step 5 Preview vs Gemini 4 Argon
- Claude Opus 4.7 vs Step 5 Preview
- Claude Opus 4.8 vs GPT-6.1 Sol
- Claude Opus 4.8 vs GPT-5.6 Sol
Rankings: Best LLM overall · Best LLM for agents · Best cheap LLMs
Claude Opus 4.8 vs Step 5 Preview: questions
- Is Claude Opus 4.8 better than Step 5 Preview?
- Claude Opus 4.8 (high) and Step 5 Preview are neck and neck on quality (60.4 vs 54.3, within the ±7.2 margin). The BenchLeader Index combines every independent quality benchmark; the two are within the margin of error as of 13 Nov 2026, so check the category scores for your use.
- Claude Opus 4.8 vs Step 5 Preview: which is cheaper, and will that change?
- Step 5 Preview is cheaper today: $1.43 against $10.00 per million tokens, blended at three input tokens per output token. No price change has been announced for either.
- Is Claude Opus 4.8 or Step 5 Preview better for coding?
- Claude Opus 4.8 scores higher in coding (61 vs 50 on the category index, where 50 is average). On Arena Coding, the top shared coding benchmark, it is 1534 for Claude Opus 4.8 against 1505 for Step 5 Preview.
- Which is faster, Claude Opus 4.8 or Step 5 Preview?
- Step 5 Preview streams ~2× faster (135 vs 68 output tokens per second, Artificial Analysis 7-day median).
- Which is better for long documents?
- Both accept 1,000,000 tokens of context.
More questions (3)
- Which has lower latency?
- Claude Opus 4.8 answers first: 0.8 s against 22.8 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 Claude Opus 4.8 vs Step 5 Preview?
- Across the 8 benchmarks both have, Claude Opus 4.8 leads 8, Step 5 Preview leads 0. Every overall index that rates both puts Claude Opus 4.8 ahead.
- When was Step 5 Preview released?
- Step 5 Preview was released on 18 Sep 2026.