# Claude Haiku 4.5 vs Step 5 Preview

> Side-by-side benchmark comparison for Claude Haiku 4.5 and Step 5 Preview across agentic, coding, multimodal, knowledge, reasoning, multilingual, and math tasks.

- Canonical page: https://benchlm.ai/compare/claude-haiku-4-5-vs-step-5-preview
- Last updated: September 21, 2026

- Shared sourced benchmarks: 1
- HTML indexing: indexable
- Ranking lane: BenchAlign v5

## Quick Verdict

Evidence is too thin for a benchmark verdict: Claude Haiku 4.5 and Step 5 Preview share only 1 sourced benchmark row.

## Summary

- Only 1 shared sourced benchmark row is available. That is not enough for a broad capability verdict, so this mirror does not name an overall winner.
- Step 5 Preview has the lower published output-token price.
- Step 5 Preview has the larger published context window at 1M.

## Model Snapshot

| Property | Claude Haiku 4.5 | Step 5 Preview |
|----------|----------|----------|
| Creator | Anthropic | StepFun |
| Type | Proprietary | Pending |
| Reasoning | Non-Reasoning | Reasoning |
| Context | 200K | 1M |
| Independent public score (not head-to-head) | 53.17 | null |
| Benchmarks Covered | 10 | 22 |
| Pricing (input/output) | $1.00 / $5.00 | $1.00 / $2.70 |

## Category Breakdown

### Agentic

- Winner: Insufficient shared evidence
- Claude Haiku 4.5 public-lane score: 28.8 (Supported · #147/154)
- Step 5 Preview public-lane score: Coming soon

| Benchmark | Claude Haiku 4.5 | Step 5 Preview | Winner |
|-----------|-----------|-----------|--------|
| JobBench | 16.0% | 59.0% | Step 5 Preview |
| Terminal-Bench 2.1 (Vals) | 43.8% | Coming soon | Coming soon |
| Terminal-Bench 2.1 | Coming soon | 85.0% | Coming soon |
| Terminal-Bench 4.0 | Coming soon | 33.30% | Coming soon |
| CyberGym | Coming soon | 84.7% | Coming soon |
| AutomationBench | Coming soon | 44.0% | Coming soon |
| Toolathlon-Verified | Coming soon | 74.1% | Coming soon |
| MCP Atlas | Coming soon | 85.6% | Coming soon |
| APEX-Agents | Coming soon | 37.8% | Coming soon |
| DRACO | Coming soon | 83.3% | Coming soon |
| BrowseComp | Coming soon | 88.7% | Coming soon |
| Agents' Last Exam | Coming soon | 29.5% | Coming soon |

### Coding

- Winner: Insufficient shared evidence
- Claude Haiku 4.5 public-lane score: 27.9 (Supported · #149/156)
- Step 5 Preview public-lane score: Coming soon

| Benchmark | Claude Haiku 4.5 | Step 5 Preview | Winner |
|-----------|-----------|-----------|--------|
| SWE-bench Verified | 73.3% | Coming soon | Coming soon |
| VulcanBench v3 | 76.2% | Coming soon | Coming soon |
| LiveCodeBench (Vals) | 41.2% | Coming soon | Coming soon |
| SWE-bench (Vals) | 66.6% | Coming soon | Coming soon |
| DeepSWE | Coming soon | 67.7% | Coming soon |
| Terminal-Bench 2.1 | Coming soon | 85.0% | Coming soon |
| SciCode | Coming soon | 58.9% | Coming soon |
| ProgramBench | Coming soon | 80.5% | Coming soon |
| sweMarathon | Coming soon | 72.7% | Coming soon |
| MLS-Bench Lite | Coming soon | 40.5% | Coming soon |

### Reasoning

- Winner: Insufficient shared evidence
- Claude Haiku 4.5 public-lane score: Coming soon
- Step 5 Preview public-lane score: 78.5 (Unranked · 2 rankable rows)

| Benchmark | Claude Haiku 4.5 | Step 5 Preview | Winner |
|-----------|-----------|-----------|--------|
| CritPt | Coming soon | 20.9% | Coming soon |

### Multimodal & Grounded

- Winner: Insufficient shared evidence
- Claude Haiku 4.5 public-lane score: Coming soon
- Step 5 Preview public-lane score: 61.2 (#28/49)

| Benchmark | Claude Haiku 4.5 | Step 5 Preview | Winner |
|-----------|-----------|-----------|--------|
| OfficeQA Pro | Coming soon | 60.3% | Coming soon |
| MMMU-Pro | Coming soon | 76% | Coming soon |

### Knowledge

- Winner: Insufficient shared evidence
- Claude Haiku 4.5 public-lane score: 44 (Estimated · #120/186)
- Step 5 Preview public-lane score: Coming soon

| Benchmark | Claude Haiku 4.5 | Step 5 Preview | Winner |
|-----------|-----------|-----------|--------|
| GPQA Diamond (Vals) | 72.2% | Coming soon | Coming soon |
| MMLU-Pro (Vals) | 78.7% | Coming soon | Coming soon |
| GPQA-D | Coming soon | 93.5% | Coming soon |
| HLE | Coming soon | 46.5% | Coming soon |

### Mathematics

- Winner: Insufficient shared evidence
- Claude Haiku 4.5 public-lane score: 28.8 (Unranked · 2 rankable rows)
- Step 5 Preview public-lane score: Coming soon

| Benchmark | Claude Haiku 4.5 | Step 5 Preview | Winner |
|-----------|-----------|-----------|--------|
| FrontierMath v2 (Tiers 1-3) | 5.903% | Coming soon | Coming soon |
| FrontierMath v2 (Tier 4) | 2.083% | Coming soon | Coming soon |

## FAQ

### Is there enough evidence to pick between Claude Haiku 4.5 and Step 5 Preview?

Not yet. The models share only 1 sourced benchmark row, so this mirror reports the available measurements and metadata without naming a broad winner.

### What can I compare safely on this page?

Published pricing, context windows, model type, and individual shared benchmark rows are directly comparable. Independent overall scores and non-overlapping benchmark coverage are labeled separately and should not be read as a controlled head-to-head result.

## Related Comparisons

- [Claude Haiku 4.5 vs Claude Haiku 4.5 Thinking](/compare/claude-haiku-4-5-vs-claude-haiku-4-5-thinking)
- [Step 5 Preview vs Claude Haiku 4.5 Thinking](/compare/claude-haiku-4-5-thinking-vs-step-5-preview)
- [Claude Haiku 4.5 vs Claude Fable 5.1](/compare/claude-fable-5-1-vs-claude-haiku-4-5)
- [Step 5 Preview vs Claude Fable 5.1](/compare/claude-fable-5-1-vs-step-5-preview)

## Explore More

- [Claude Haiku 4.5 profile](/models/claude-haiku-4-5)
- [Step 5 Preview profile](/models/step-5-preview)
- [Compare Pricing](/llm-pricing)
- [Alternative Finder](/tools/alternative-finder)
- [LLM Selector](/tools/llm-selector)
- [Overall Rankings](/best/overall)