Anthropic: Claude Sonnet 5 vs StepFun: Step 3.7 Flash
Side-by-side comparison of specs, pricing, benchmark scores, and task rankings. Updated 2026-09-29.
| Anthropic: Claude Sonnet 5 | StepFun: Step 3.7 Flash | |
|---|---|---|
| Vendor | anthropic | stepfun |
| Quality Score | 100 | 100 |
| Benchmark Score | 63.9 | 29.1 |
| Input Price | $2.00/M | $0.20/M |
| Output Price | $10.00/M | $1.15/M |
| Context Window | 1,000,000 | 262,144 |
| Max Output | 128,000 | 230,400 |
| Tool Calling | ✓ | ✓ |
| Structured Output | ✓ | ✓ |
| Reasoning Mode | ✓ | ✓ |
| Vision | ✓ | ✓ |
| Audio | - | - |
| Benchmark Scores | ||
| ai_index | 63.0 | 32.1 |
The verdict
For a sample job of 1 million input tokens and 250,000 output tokens, Step 3.7 Flash costs $0.49 and Claude Sonnet 5 costs $4.50, so Step 3.7 Flash is about 9.2 times cheaper for the same work.
Claude Sonnet 5 reads up to 1,000,000 tokens at once, roughly 1,500 pages of text, against about 393 pages for Step 3.7 Flash.
Only Step 3.7 Flash offers video input.
On our blended benchmark score Claude Sonnet 5 leads 63.9 to 29.1.
Across the 20 tasks where they differ, Claude Sonnet 5 scores higher on 16 (SQL Generation, Code Review, Code Refactoring and others), Step 3.7 Flash on 4 (Code Completion, JSON Extraction, Bulk Data Labeling and others).
Claude Sonnet 5 was released in June 2026 and Step 3.7 Flash in May 2026.
Pick Claude Sonnet 5 if you need long documents, benchmark performance or the stronger all-round task scores; pick Step 3.7 Flash if you need lower cost or video input.
Costs use current list prices per million tokens; page counts assume about 0.75 words per token and 500 words per page.
Who wins by task?
| Task | Anthropic: Claude Sonnet 5 | StepFun: Step 3.7 Flash |
|---|---|---|
| SQL Generation | 140 | 135 |
| Code Review | 144 | 132 |
| Code Completion | 118 | 129 |
| Code Refactoring | 146 | 132 |
| Bug Fixing | 148 | 136 |
| Unit Test Generation | 131 | 125 |
| Code Documentation | 135 | 129 |
| Regex Writing | 123 | 122 |
| CI/CD Pipelines | 127 | 121 |
| Frontend Component Design | 129 | 125 |
| Data Analysis | 133 | 128 |
| CSV / Spreadsheet Cleanup | 135 | 129 |
| ETL Scripting | 136 | 127 |
| JSON Extraction | 121 | 131 |
| Bulk Data Labeling | 118 | 129 |
| OCR / Document Parsing | 134 | 129 |
| Table Extraction from PDFs | 134 | 129 |
| Long-Document Summarization | 146 | 134 |
| Short-Form Summarization | 116 | 125 |
| Blog Post Writing | 128 | 123 |
Scores reflect capability match + benchmark data + pricing for each task. Methodology →