Loading benchmark catalogue
Compare performance benchmarks, coding intelligence, reasoning evaluations, context limits, and API economics between Claude Opus 4.8 (Anthropic) and Step 5 Preview (StepFun).
| Metric / Capability | Claude Opus 4.8 | Step 5 Preview |
|---|---|---|
| Developer / Organization | Anthropic | StepFun |
| Release Date | 2026-05-27 | 2026-10-08 |
| Deployment / Weights | Proprietary Cloud API | Open Weights |
| Context Window | 1M tokens (1,000,000 tokens) | 1.0M tokens (1,048,576 tokens) |
| Input Pricing (/1M Tokens) | $5 | Free (Open Weights) |
| Output Pricing (/1M Tokens) | $25 | Free (Open Weights) |
| Benchmark / Domain | Claude Opus 4.8 | Step 5 Preview | Advantage |
|---|---|---|---|
Artificial Analysis Intelligence Index Composite index aggregating 10+ evaluations of general reasoning and knowledge. | 41.8% | 43.7% | Step 5 Preview (+5%) |
Artificial Analysis Coding Index Composite index of coding and software-engineering evaluations. | 74.3% | — | — |
Artificial Analysis Agentic Index Composite index of tool-use and long-horizon agentic tasks. | 41.9% | — | — |
Artificial Analysis Coding Agent Index v2.0 Composite index of coding-agent systems (model + harness) across FrontierCode v1.1, SWE-Atlas-QnA, and Terminal-Bench v4. | — | 43.1% | — |
SWE-bench Pro Harder, contamination-resistant successor to SWE-bench spanning multiple languages. | 69.2% | — | — |
Terminal-Bench 4.0 Sandboxed command-line tasks requiring shell and file-system operation. | 23.6% | 33.3% | Step 5 Preview (+41%) |
ARC-AGI-2 Novel visual grid puzzles requiring abstract rule induction. | 72.1% | — | — |
GPQA Diamond Graduate-level physics, chemistry, and biology questions written by domain experts. | 93.6% | 93.5% | Claude Opus 4.8 leads |
Humanity's Last Exam Expert-written questions across hundreds of disciplines designed to resist saturation. | 49.8% | 46.5% | Claude Opus 4.8 (+7%) |
FrontierMath v2 (Tiers 1-3) Unpublished, research-level mathematics problems across Tiers 1-3. | 80.0% | — | — |
Design Arena ELO Human-preference ELO across UI, 3D, data-visualization, and game-dev generations. | 1310 | — | — |
Both models perform neck-and-neck across available benchmarks, with Claude Opus 4.8 and Step 5 Preview each excelling in distinct tasks.
Step 5 Preview has the larger context capacity of 1.0M tokens (1,048,576 tokens), compared to 1M tokens for Claude Opus 4.8.
Step 5 Preview is more economical. Input tokens cost $Free (Open Weights)/M tokens compared to $5/M tokens for Claude Opus 4.8.
Claude Opus 4.8 is a proprietary cloud API. Step 5 Preview is an open-weights model.
Analyze 130+ models side-by-side using interactive radar charts, cost-efficiency quadrant plots, and live dataset filters on the full NeuBench index.
Open Interactive Comparator