DeepSeek V4 Flash Official Release: A Step-by-Step Guide to Configuring It in Codex

DeepSeek V4 Flash officially launched its public beta on July 31, 2026, with native support for the Responses API and tailored compatibility for Codex. The model retains its name "deepseek-v4-flash", meaning existing API calls don’t require a name change. DeepSeek also shared vendor-provided benchmark results—including an 82.7 score on Terminal Bench 2.1 and 70.3 on Toolathlon verified—to help users understand the model’s positioning, though these figures shouldn’t be taken as direct indicators of local development experiences.

Before setting up DeepSeek V4 Flash in Codex, ensure you have three prerequisites: a Codex CLI or client (verify via codex --version), an existing Codex installation (confirm the ~/.codex directory exists), and a DeepSeek API Key (created from the DeepSeek Platform).

For Windows users, the setup is a one-line command in PowerShell: irm https://cdn.deepseek.com/api-docs/codex-deepseek-setup-en.ps1 | iex. After running the script, input "1" and follow prompts to enter your API Key. The script will back up your original config.toml, update models.json, set deepseek-v4-flash as the default model, and point the provider to DeepSeek’s Responses API. Note: During testing, a separate CODEX_HOME was used to avoid modifying personal ~/.codex settings.

For macOS or Linux, use the bash command: bash <(curl -fsSL https://cdn.deepseek.com/api-docs/codex-deepseek-setup.sh). The configuration file is shared across Codex CLI, desktop client, and VS Code plugin—restart any running clients to avoid using outdated cache.

To verify if the configuration works, Windows desktop users may see "custom" at the bottom left, but this alone doesn’t confirm API connectivity. Use codex doctor --json to check four key metrics: config.load should be "OK", the model should be "deepseek-v4-flash", the provider should be "deepseek", and the wire API should be "responses". A successful check means Codex recognizes the configuration, though a real task with a valid API Key is needed to confirm full connectivity.

A common compatibility issue arises with Codex CLI 0.137.0: the official script generates models.json with an "effort" value of "max", but Codex’s parser only accepts "none", "minimal", "low", "medium", "high", or "xhigh", leading to an "unknown variant max" error. To fix this, run the following PowerShell command: $path = "$HOME\.codex\models.json"; $json = [IO.File]::ReadAllText($path); $json = $json.Replace('\"effort\": \"max\"', '\"effort\": \"xhigh\"'); [IO.File]::WriteAllText($path, $json, [Text.UTF8Encoding]::new($false)). Only apply this fix if you encounter the error—updating Codex may resolve it without manual changes.

API key management is critical: avoid committing config.toml, terminal history, or screenshots with keys to Git or public platforms. For teams, assign individual keys so leaks can be revoked without affecting others. To roll back to the original configuration, re-run the official script and select option "3"—this uses the backup in ~/.codex/backup-deepseek. If you made manual changes, check the backup time first to avoid overwriting recent settings.

When using Codex for Agent tasks, don’t rely solely on API标价—long context and frequent tool calls can lead to higher costs than expected. Compare DeepSeek V4’s different channels on price, latency, and availability, then test the same task across channels to check token consumption. This helps choose the most cost-effective line.

Sources

• Juejin Tech Community: https://juejin.cn/post/7669638812705587238