DeepSeek · Released Jul 31, 2026DeepSeek V4 Flash 0731
Official DeepSeek V4 Flash release with enhanced agentic capabilities and integrated DSpark speculative decoding.
- Capability
- 154.5ECI · #32 of 148
- Input
- $0.14per 1M tokens
- Output
- $0.28per 1M tokens
- Context
- 1M384K max output
What it costs
- Input
- $0.14per million tokens
- Output
- $0.28per million tokens
- Blended (3:1)
- $0.175cheaper than 79% of priced models
Typical monthly bills
| Monthly usage | Estimated cost |
|---|---|
| Side project2M input + 0.5M output tokens | $0.42 |
| Team assistant25M input + 5M output tokens | $4.90 |
| Production app250M input + 50M output tokens | $49.00 |
Median of 48 paid API providers on models.dev.
Independent benchmarks
#32 of 148 scored models
The shaded band is Epoch AI’s confidence range (152.0–156.6); the tick marks the median scored model.
Scores from Epoch AI, run independently of DeepSeek.
The details
- Lab
- DeepSeek
- Released
- Jul 31, 2026
- Knowledge cutoff
- May 2025
- Context window
- 1,000,000 tokens
- Max output
- 384,000 tokens
- Inputs
- Text
- Output
- Text
- Reasoning
- Always on
- Tool calling
- Yes
- Structured output
- Yes
- Weights
- Open weights · MIT
- Availability
- 49 API providerslisted on models.dev
What DeepSeek claims
Published by the lab at launch. Settings vary, so compare these only with care.
| Benchmark | Score | Setting | Source |
|---|---|---|---|
| Terminal-Bench v2.1 | 82.7 | max | Source |
| NL2Repo | 54.2 resolve rate | max effort | Source |
| CyberGym | 76.7 | max effort | Source |
| DeepSWE | 54.4 resolve rate | max effort | Source |
| Toolathlon-Verified | 70.3 | max effort | Source |
| Agents' Last Exam | 25.2 | max effort | Source |
| AutomationBench | 25.1 success rate | max effort | Source |
| DSBench-FullStack | 68.7 | max effort | Source |
| DSBench-Hard | 59.6 | max effort | Source |
More from DeepSeek
About DeepSeek V4 Flash 0731
How much does DeepSeek V4 Flash 0731 cost?
DeepSeek V4 Flash 0731 costs $0.14 input / $0.28 output per million tokens (median across 48 API providers). At a 3:1 input-to-output mix that is $0.175 per million tokens, cheaper than 79% of the 360 priced models we track.
What is the context window of DeepSeek V4 Flash 0731?
DeepSeek V4 Flash 0731 accepts up to 1,000,000 tokens per request and can write up to 384,000 tokens in one response.
How good is DeepSeek V4 Flash 0731?
Epoch AI gives DeepSeek V4 Flash 0731 a Capabilities Index score of 154.5 (likely range 152.0–156.6), ranking it #32 of 148 models Epoch has scored. Epoch AI benchmark results: GPQA Diamond 91.0%, FrontierMath Tiers 1–3 57.5%, OTIS Mock AIME 2024–2025 94.4%, SimpleQA Verified 33.6%.
Is DeepSeek V4 Flash 0731 open source?
Yes. DeepSeek publishes the weights under the MIT licence, so it can be downloaded and self-hosted.
What inputs does DeepSeek V4 Flash 0731 support?
DeepSeek V4 Flash 0731 accepts text and replies in text. It is a reasoning model, supports tool calling and can return structured JSON output.
When was DeepSeek V4 Flash 0731 released?
DeepSeek released DeepSeek V4 Flash 0731 on Jul 31, 2026. Its training data runs to May 2025.
What are the best alternatives to DeepSeek V4 Flash 0731?
The closest current models from other labs on capability, price and release date are GPT-6 Luna (OpenAI, $0.10 / $0.50), GLM-5.3-Flash (Z.ai (Zhipu), ECI 151.9, $0.15 / $0.50), Gemini 3.6 Flash (Google, ECI 154.3, $0.75 / $3.75) and Inkling Small (Thinking Machines, ECI 150.2, $0.50 / $1.20).