V4 Flash 0731
DeepSeek · released Jul 31, 2026
DeepSeek V4 Flash 0731 is a sparse mixture-of-experts model from DeepSeek, with 13B active parameters out of 284B total. This re-post-trained revision is suited for coding, reasoning, and agent workflows.
Specification
- Context window
- 1.05M
- Max output
- 66K
- Knowledge cutoff
- Not stated
- Parameters
- 304.2B
- Licence
- mit
- Serving providers
- 23
- Moderated
- No
- Uptime
- 100.0%
Intelligence
51.8
96th percentile
Coding
69.1
Coding Index
Agentic
48.4
Agentic Index
Output speed
148 t/s
Median across providers
Latency
324ms
Time to first token
Cost per task
$0.02
Estimated
Benchmarks
Where the score comes from
The Intelligence Index is a composite. These are the underlying evaluations this model was actually measured on.
Evaluation scores
Percentage correct · higher is better
- GPQA Diamond90.8%
- AA-LCR (long context)74.3%
- SciCode49.9%
- τ²-bench (Telecom)39.4%
- Humanity's Last Exam38.6%
An evaluation missing from this list was not run for this model — it is not a zero.
View as table
| Evaluation | Score |
|---|---|
| GPQA Diamond | 90.8% |
| AA-LCR (long context) | 74.3% |
| SciCode | 49.9% |
| τ²-bench (Telecom) | 39.4% |
| Humanity's Last Exam | 38.6% |
Against its peers
Intelligence Index · this model highlighted, nearest peers in grey
- Muse Spark 1.1Meta53.2
- GPT-5.4OpenAI53.1
- GLM 5.2Z.ai52.6
- GPT-5.6 LunaOpenAI52.3
- Gemini 3.5 FlashGoogle52.0
- V4 Flash 0731DeepSeek51.8
- V4 Flash 0423DeepSeek51.8
- Gemini 3.6 FlashGoogle51.6
Peers are the models sitting closest on the Intelligence Index — the set you would realistically choose between.
View as table
| Model | Intelligence |
|---|---|
| Muse Spark 1.1 | 53.2 |
| GPT-5.4 | 53.1 |
| GLM 5.2 | 52.6 |
| GPT-5.6 Luna | 52.3 |
| Gemini 3.5 Flash | 52.0 |
| V4 Flash 0731 | 51.8 |
| V4 Flash 0423 | 51.8 |
| Gemini 3.6 Flash | 51.6 |
Percentile among all indexed models
Pricing
What it costs to run
List prices per million tokens, plus what one representative task works out to.
List price
- Input / 1M tokens
- $0.09
- Output / 1M tokens
- $0.18
- Cached input / 1M
- $0.018
- Blended 3:1
- $0.113
One task, estimated
$0.02
- Input tokens
- 50,000
- Output tokens
- 80,000
- Profile
- Reasoning
Estimated from list pricing: 50K input tokens plus 80K output tokens for reasoning models (25K for non-reasoning).
Serving providers
- DeepInfra
- BaseTen
- CoreWeave
- Morph
- Fireworks
- Baidu
- DigitalOcean
- AkashML
- Novita
- Cloudflare
- Together
- DeepSeek
- StreamLake
- Parasail
- AtlasCloud
- GMICloud
- SiliconFlow
- Ionstream
- Ambient
- Io Net
- Venice
- Mancer 2
- Phala
Speed and latency figures are medians across these providers, so a widely-served model reports a blend rather than any single endpoint.
Arena
Head-to-head generation quality
Elo from pairwise judgements, broken out by the kind of thing the model was asked to build.
- Overall Elo
- 1,279
- Win rate
- 53.6%
- Strongest at
- 3D scenes
- Tournaments
- 528
Elo by category
Dot position on a 1,175–1,300 scale · Elo has no meaningful zero
- 1,279
- 1,264
- 1,259
- 1,248
- 1,196
Agent categories (full-stack apps, mobile apps) are only scored for models tested in agent harnesses.
View as table
| Category | Elo | Win rate |
|---|---|---|
| 3D scenes | 1279 | 53.6% |
| UI components | 1264 | 47.0% |
| Websites | 1259 | 47.0% |
| Game development | 1248 | 45.6% |
| Data visualisation | 1196 | 40.7% |