Step 3.7 Flash vs DeepSeek V4-Flash
Step 3.7 Flash is from StepFun, released May 29, 2026. DeepSeek V4-Flash is from DeepSeek, released July 31, 2026.
- Input per 1M$0.20 vs $0.15. DeepSeek V4-Flash costs 25% less per million input tokens.
- Output per 1M$1.15 vs $0.60. DeepSeek V4-Flash costs 48% less per million output tokens.
- Context window256k vs 1M. DeepSeek V4-Flash takes 291% more context in one request.
Prices are the vendors’ published API rates; Elo is LMArena’s text board. Each figure is dated on its model page.
Interactive Comparison
Loading chart...
$0.20 / $1.15 per million tokens | $0.15 / $0.60 per million tokens | |
|---|---|---|
| Identity | ||
| Developer | StepFun | DeepSeek |
| Released | May 2026 | Jul 2026 |
| Retired | — | 10 September 2026 |
| Status | — | Retired. DeepSeek retired V4-Flash on 2026-09-10; requests to deepseek-v4-flash are now served by DeepSeek V4.1-Flash, which has its own page. The figures below are V4.1-Flash's (off-peak), recorded here before that page existed. |
| Licence | Apache 2.0 (open weights) | MIT |
| Self-hostable | Yes — stepfun-ai/Step-3.7-Flash on Hugging Face | Yes - open weights on Hugging Face |
| Cost | ||
| Blended $/1M tokensinput × 0.75 + output × 0.25 | $0.438 / 1M tokens | $0.262 / 1M tokens |
| Input price | $0.200 / 1M tokens | $0.150 / 1M tokens |
| Output price | $1.150 / 1M tokens | $0.600 / 1M tokens |
| Cached input | $0.04 / 1M tokens | $0.006 / 1M tokens (peak); $0.003 off-peak - V4.1-Flash rates, which now serve this name |
| Batch discount | — | Off-peak billing: 50% off all rates outside 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday (weekends and Chinese public holidays are off-peak in full) |
| Free tier | — | Not offered |
| Capacity | ||
| Context window | 256K tokens | 1M tokens |
| Max output | — | 384k tokens |
| Long-context surcharge | — | None — flat rate across the 1M window |
| Capability | ||
| Vision in | Yes — images and video | No |
| Audio in | — | No |
| Function calling | Yes | Yes |
| Structured output | — | Yes — JSON output |
| Extended reasoning | Yes — three reasoning-effort levels | Yes — thinking (default) and non-thinking modes |
| Web search | — | No |
| Code execution | — | No |
| Access | ||
| API | Yes — StepFun Open Platform (step-3.7-flash) | Yes |
| Chat app | — | Yes |
| Cloud marketplaces | — | Not offered |
| Fine-tuning | — | Not offered |
| Measured quality | ||
| LMArena Elo | — | 1436 (checked Oct 2026) |
| LMArena rank | — | Rank 105 |
| Elo per dollarLMArena Elo ÷ blended $/1M tokens | — | 5470 |
