Is DeepSeek V4-Flash 0731 a brand-new model?

DeepSeek V4-Flash 0731 keeps the same 284B mixture-of-experts architecture as V4-Flash-preview — the team only re-ran post-training. The change targets agent behaviour, and the checkpoint now scores 82.7 on Terminal Bench 2.1, 76.7 on Cybergym, and 54.4 on DeepSWE.

How much does DeepSeek V4-Flash 0731 cost?

DeepSeek V4-Flash costs $0.14 per million input tokens on a cache miss, $0.0028 per million on a cache hit, and $0.28 per million output tokens. Pricing is unchanged from the preview and remains among the cheapest ways to call a top-lab open-weight model at agent quality.

Does DeepSeek V4-Flash 0731 work with the Codex CLI?

DeepSeek V4-Flash 0731 natively speaks the Responses API format and is tuned for the Codex adapter, so tools that expect the OpenAI Responses interface can slot it in without a shim. The API endpoint stays the same — set the model name to `deepseek-v4-flash`.

What did DeepSeek use to score the new numbers?

DeepSeek benchmarked V4-Flash 0731 with the DeepSeek Harness in minimal mode, max effort level, top-p 0.95, and temperature 1.0. DeepSeek says the harness will be released soon, so third parties can reproduce the scores themselves.

Does this update also change V4-Pro?

No — the 0731 update only refreshes the V4-Flash API. DeepSeek V4-Pro and the app/web models are unchanged. Teams already using V4-Pro or the DeepSeek app will not see any behaviour shift from this release.

DeepSeek · 2026-07-31 · major

DeepSeek V4-Flash goes official — 0731 hits 82.7 on Terminal Bench 2.1

DeepSeek V4-Flash exits preview with a re-post-trained checkpoint on the same 284B MoE, hitting 82.7 on Terminal Bench 2.1, 76.7 on Cybergym, and 68.7 on DSBench-FullStack. Input pricing stays at $0.14 per million tokens.

DeepSeek V4-Flash model card social thumbnail on Hugging Face

DeepSeek graduates V4-Flash from preview to official with a re-post-trained checkpoint tuned for agents.

Key specs

Terminal bench 2.1	82.7
Dsbench full stack	68.7

Quick facts

Maker	DeepSeek
Model name	deepseek-v4-flash-0731
Architecture	284B MoE, 13B active
Input price	$0.14 / 1M tokens
Output price	$0.28 / 1M tokens
Availability	API public beta
License	MIT (open weights)

Pricing

Input (cache miss)	$0.14 / 1M tokens
Input (cache hit)	$0.0028 / 1M tokens
Output	$0.28 / 1M tokens

source ↗

What is it?

DeepSeek V4-Flash 0731 is the official public-beta release of DeepSeek's 284B mixture-of-experts model, replacing the earlier V4-Flash-preview API endpoint. It shares the same architecture and parameter count as the preview but ships a re-post-trained checkpoint focused on agent workloads.

How does it work?

The 0731 checkpoint keeps V4-Flash's 284B/13B-active MoE unchanged and re-runs only the post-training pass on agent-flavoured data. Responses API support is native, and Codex compatibility is baked in so tools that expect OpenAI's Responses interface work without a shim.

Why does it matter?

Open-weight agent models keep closing the gap with closed-source options, and V4-Flash at $0.14 in / $0.28 out per million tokens is a fraction of most agent-tier APIs. The jump to 82.7 on Terminal Bench 2.1 and 68.7 on DSBench-FullStack lets teams run harder tool-using workloads at Flash prices.

Who is it for?

coding-agent builders, cost-conscious API users

Frequently asked questions

Is DeepSeek V4-Flash 0731 a brand-new model?: DeepSeek V4-Flash 0731 keeps the same 284B mixture-of-experts architecture as V4-Flash-preview — the team only re-ran post-training. The change targets agent behaviour, and the checkpoint now scores 82.7 on Terminal Bench 2.1, 76.7 on Cybergym, and 54.4 on DeepSWE.
How much does DeepSeek V4-Flash 0731 cost?: DeepSeek V4-Flash costs $0.14 per million input tokens on a cache miss, $0.0028 per million on a cache hit, and $0.28 per million output tokens. Pricing is unchanged from the preview and remains among the cheapest ways to call a top-lab open-weight model at agent quality.
Does DeepSeek V4-Flash 0731 work with the Codex CLI?: DeepSeek V4-Flash 0731 natively speaks the Responses API format and is tuned for the Codex adapter, so tools that expect the OpenAI Responses interface can slot it in without a shim. The API endpoint stays the same — set the model name to `deepseek-v4-flash`.
What did DeepSeek use to score the new numbers?: DeepSeek benchmarked V4-Flash 0731 with the DeepSeek Harness in minimal mode, max effort level, top-p 0.95, and temperature 1.0. DeepSeek says the harness will be released soon, so third parties can reproduce the scores themselves.
Does this update also change V4-Pro?: No — the 0731 update only refreshes the V4-Flash API. DeepSeek V4-Pro and the app/web models are unchanged. Teams already using V4-Pro or the DeepSeek app will not see any behaviour shift from this release.

Try it

Set `model: 'deepseek-v4-flash'` against `https://api.deepseek.com`.