What is DeepSeek V4 Flash 0731?

DeepSeek-V4-Flash-0731 is the official DeepSeek V4 Flash build released on 31 July 2026. It keeps the same architecture and parameter count as the preview build — only the post-training was redone. The gains are concentrated in agentic and coding work: Terminal Bench 2.1 rose from 61.8 to 82.7, DeepSWE from 7.3 to 54.4, and DSBench-FullStack from 37.0 to 68.7.

Do I need to change my model ID to get the 0731 build?

No. DeepSeek kept the model ID unchanged — calling deepseek-v4-flash serves the latest 0731 build automatically. Existing OpenClaw agents already using DeepSeek V4 Flash are on the new build with no redeploy or config change.

How much does DeepSeek V4 Flash cost?

On the direct DeepSeek API, V4 Flash is $0.14 per million input tokens on a cache miss, $0.0028 on a cache hit, and $0.28 per million output tokens. Through OpenRouter it is roughly $0.09 input and $0.18 output per million tokens. OpenClaw Launch hosting starts at $3/month with AI credits included.

Is DeepSeek V4 Flash better than V4 Pro?

On DeepSeek's published agent benchmarks, the 0731 Flash build matches or beats the V4-Pro-Preview on several tests — it scores 82.7 on Terminal Bench 2.1 versus 72.1 for V4-Pro-Preview — while costing about a third as much per output token. V4 Pro remains the stronger pick for the hardest long-horizon reasoning and deep multi-file refactors.

What context window does DeepSeek V4 Flash have?

DeepSeek V4 Flash supports a 1 million token context window with a maximum output length of 384K tokens.

← Home

Guide

OpenClaw + DeepSeek V4 Flash: Setup, 0731 Benchmarks, and Pricing

Q: Does DeepSeek V4 Flash support the Responses API and Codex?

Yes. The 0731 build natively supports the Responses API request format and is advertised as compatible with Codex-style clients, so tooling written against that shape can call V4 Flash directly.

DeepSeek refreshed V4 Flash on 31 July 2026. Same architecture, same parameter count, rebuilt post-training — and a large jump on agent and coding benchmarks. Here is what changed and how to run it on OpenClaw.

What Changed in the 0731 Build

DeepSeek updated deepseek-v4-flash to the DeepSeek-V4-Flash-0731 build. This is not a new model generation. DeepSeek has been explicit that the architecture and parameter count are unchanged from the preview build — it is a post-training upgrade, and the gains land almost entirely on agentic and coding work.

The practical detail that matters most: the model ID did not change. Per DeepSeek's release note, you still call deepseek-v4-flash and you are served the latest build. If you already run a DeepSeek V4 Flash agent on OpenClaw Launch, you are on the 0731 build already — no redeploy, no config edit, no model switch.

DeepSeek V4 Flash 0731 Benchmarks

DeepSeek published nine agent-focused benchmark results for the official build. The three with published preview-build comparisons show how lopsided the retraining was:

Benchmark	Preview build	0731 build
Terminal Bench 2.1	61.8	82.7
DeepSWE	7.3	54.4
DSBench-FullStack	37.0	68.7

The remaining six published scores for the 0731 build:

Benchmark	Score
Cybergym	76.7
Toolathlon (verified)	70.3
DSBench-Hard	59.6
NL2Repo	54.2
Agent Last Exam	25.2
Automation Bench (Public)	25.1

For scale: V4-Pro-Preview scores 72.1 on Terminal Bench 2.1, so the cheap Flash model now runs ahead of the previous flagship preview on that test. DeepSeek reports the official Flash build matching or beating V4-Pro-Preview across several of the nine benchmarks.

Read this honestly: these are DeepSeek's own published numbers on agent-flavored benchmarks, not independent evaluations, and a jump from 7.3 to 54.4 on DeepSWE says as much about how weak the preview build was on that test as about how strong 0731 is. Frontier models like Claude Opus 5 and GPT-5.6 still lead on the hardest long-horizon coding work. What the 0731 build genuinely changes is the price-to-capability floor for routine agent loops.

Responses API and Codex Compatibility

The 0731 build natively supports the Responses API format and is advertised as compatible with Codex-style clients. If you have tooling written against that request shape, V4 Flash can now sit behind it directly rather than through a translation layer.

DeepSeek V4 Flash Pricing

Flash is priced well below Pro, and DeepSeek's cache-hit rate makes repeated-context agent loops extremely cheap. Direct DeepSeek API pricing per million tokens:

Model	Input (cache hit)	Input (cache miss)	Output	Concurrency
DeepSeek V4 Flash	$0.0028	$0.14	$0.28	2,500
DeepSeek V4 Pro	$0.003625	$0.435	$0.87	500

Routed through OpenRouter instead, V4 Flash runs about $0.09 input / $0.18 output per million tokens. Two things to plan around: DeepSeek has said direct-API pricing will move to a peak/off-peak model where peak hours cost 2× (peak is 09:00–12:00 and 14:00–18:00 Beijing time), and the cache-hit rate is roughly 50× cheaper than cache-miss — so a long, stable system prompt reused across turns is worth structuring for.

Deploy DeepSeek V4 Flash on OpenClaw Launch

The fastest path is the managed route — no DeepSeek account and no API key, billed against your included AI credits.

Go to openclawlaunch.com and open the configurator.
Pick DeepSeek V4 Flash in the model dropdown.
Choose your chat platform — Telegram, Discord, WhatsApp, WeChat, or the browser gateway.
Click Deploy. Your agent is live in about 30 seconds.

Already running V4 Flash? You are on the 0731 build automatically, because the model ID is unchanged upstream. Nothing to do.

Self-Hosted Configuration

Running your own OpenClaw instance? Point the default agent at V4 Flash through OpenRouter in openclaw.json:

{
  "models": {
    "providers": {
      "openrouter": {
        "apiKey": "sk-or-..."
      }
    }
  },
  "agents": {
    "defaults": {
      "model": {
        "primary": "openrouter/deepseek/deepseek-v4-flash"
      }
    }
  }
}

Get an OpenRouter key at openrouter.ai/keys. To call DeepSeek directly instead of through OpenRouter, configure the deepseek provider and set the primary model to deepseek/deepseek-v4-flash.

Flash or Pro After the 0731 Build?

The 0731 refresh narrowed this decision considerably. Flash used to be the compromise you accepted to save money on agent loops; on DeepSeek's published agent benchmarks it now leads V4-Pro-Preview on several tests while costing roughly a third as much on output tokens.

Start with V4 Flash for agent tool loops, terminal and coding work, high-volume chat, and anything where you are paying per turn. It is the default worth beating, and at 2,500 concurrency it is the better fit for busy bots.
Reach for V4 Pro for the hardest long-horizon reasoning — deep multi-file refactors, architecture work, and analysis that genuinely walks the full 1M context.
Switch freely. Change models from the dashboard or with the /model chat command. No redeploy needed, so benchmark both on your own workload rather than trusting anyone's table.

BYOK with DeepSeek V4 Flash

Already have an OpenRouter account or a direct DeepSeek API key? Bring it to OpenClaw Launch via BYOK — select BYOK in the configurator and paste your key. Requests route through your key and your own billing.

What's Next?

DeepSeek V4 Pro vs V4 Flash — the full family comparison and how to choose
DeepSeek V4 Pro guide — setup for the flagship reasoning model
Compare all models — how V4 Flash stacks up against Claude, GPT, and Gemini
Set up Telegram — connect your DeepSeek agent to Telegram