StepFun: Step 3.7 Flash vs OpenAI: GPT-5.3-Codex
Side-by-side comparison of specs, pricing, benchmark scores, and task rankings. Updated 2026-11-07.
| StepFun: Step 3.7 Flash | OpenAI: GPT-5.3-Codex | |
|---|---|---|
| Vendor | stepfun | openai |
| Quality Score | 100 | 100 |
| Benchmark Score | 50.8 | 73.4 |
| Input Price | $0.20/M | $1.75/M |
| Output Price | $1.15/M | $14.00/M |
| Context Window | 262,144 | 400,000 |
| Max Output | 230,400 | 128,000 |
| Tool Calling | ✓ | ✓ |
| Structured Output | ✓ | ✓ |
| Reasoning Mode | ✓ | ✓ |
| Vision | ✓ | ✓ |
| Audio | - | - |
| Benchmark Scores | ||
| ai_index | 49.9 | 73.0 |
| ai_index_agentic | 35.5 | - |
| ai_index_coding | 65.3 | - |
The verdict
For a sample job of 1 million input tokens and 250,000 output tokens, Step 3.7 Flash costs $0.49 and GPT-5.3-Codex costs $5.25, so Step 3.7 Flash is about 10.8 times cheaper for the same work.
GPT-5.3-Codex reads up to 400,000 tokens at once, roughly 600 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 GPT-5.3-Codex leads 73.4 to 50.8.
Across the 20 tasks where they differ, Step 3.7 Flash scores higher on 16 (SQL Generation, Code Review, Code Completion and others), GPT-5.3-Codex on 4 (Code Refactoring, Code Documentation, ETL Scripting and others).
Step 3.7 Flash was released in May 2026 and GPT-5.3-Codex in February 2026.
Pick Step 3.7 Flash if you need lower cost, video input or the stronger all-round task scores; pick GPT-5.3-Codex if you need long documents or benchmark performance.
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 | StepFun: Step 3.7 Flash | OpenAI: GPT-5.3-Codex |
|---|---|---|
| SQL Generation | 153 | 141 |
| Code Review | 146 | 145 |
| Code Completion | 130 | 118 |
| Code Refactoring | 144 | 147 |
| Bug Fixing | 155 | 149 |
| Unit Test Generation | 139 | 132 |
| Code Documentation | 133 | 135 |
| Regex Writing | 129 | 124 |
| CI/CD Pipelines | 131 | 128 |
| Frontend Component Design | 136 | 130 |
| Data Analysis | 150 | 134 |
| CSV / Spreadsheet Cleanup | 141 | 135 |
| ETL Scripting | 137 | 138 |
| JSON Extraction | 142 | 120 |
| Bulk Data Labeling | 133 | 117 |
| OCR / Document Parsing | 138 | 134 |
| Table Extraction from PDFs | 138 | 134 |
| Long-Document Summarization | 142 | 148 |
| Short-Form Summarization | 128 | 116 |
| Blog Post Writing | 129 | 129 |
Scores reflect capability match + benchmark data + pricing for each task. Methodology →