DeepSeek V4 Flash 0731 now live in Command Code.
- DeepSeek V4 Flash now outperforms GLM 5.2
- Same model ID, Same model size
- It's re post trained
With tool calls and cache repairs in Command Code, available on all plans right away.
DeepSeek V4 Flash 0731 has significantly enhanced agent capabilities, with benchmark results far exceeding V4-Pro-Preview:
- Terminal Bench 2.1: 82.7
- NL2Repo: 54.2
- Cybergym: 76.7
- DeepSWE: 54.4
- DSBench-Hard: 59.6
- Agent Last Exam: 25.2
- Toolathlon verified: 70.3
- DSBench-FullStack: 68.7
- Automation Bench (Public): 25.1
DeepSeek V4 Flash at a glance
Fast hybrid-attention reasoning.
| Spec | DeepSeek V4 Flash |
|---|---|
| Model ID | deepseek/deepseek-v4-flash |
| Context window | 1M tokens |
| Released | July 31, 2026 |
| Intelligence Index | 34 |
| Coding Index | 69.1 |
| Input | $0.15 per 1M tokens |
| Output | $0.60 per 1M tokens |
| Cache read | $0.003 per 1M tokens |
| Effective input in an agent loop | $0.05 per 1M tokens |
| Plans | Go and above |
Specs, pricing and plans are from the DeepSeek V4 Flash model page as of September 28, 2026, including any deal running that day, so they can differ from the launch pricing above. Intelligence and Coding Index scores are from Artificial Analysis.
Use DeepSeek V4 Flash
1cmd --model deepseek/deepseek-v4-flashOr type /model in a session and pick it. You can switch mid-session without losing context.
What real work costs on DeepSeek V4 Flash
| Task | Tokens (in · out) | Cost |
|---|---|---|
| Quick lookup / one-liner | 8K · 1K | $0.0007 |
| Review a 500-line PR | 60K · 4K | $0.0052 |
| Fix a bug (agent loop) | 180K · 12K | $0.01 |
| Refactor a module | 320K · 20K | $0.03 |
| Full-repo agent run | 900K · 45K | $0.06 |
Try Command Code
The best coding agent for open models, in your terminal or on your desktop.
1npm i -g command-codeDesktop App · Docs · X · Discord
