Monday, August 3, 2026HotTea verified storyVerified 12:04 AM PDT
← Back to the Monday, August 3, 2026 edition

DeepSeek put a cheap coding-agent model into the Responses API lane.

DeepSeek's July 31 changelog says the official V4-Flash API entered public beta, keeps the `deepseek-v4-flash` model name, adds stronger agent benchmark results, natively supports the Responses API format, and is specifically adapted for Codex-style workflows. DeepSeek's Responses API guide says V4-Flash is currently the supported model for that format, with V4-Pro support expected in early August. Axios framed the release as a price-war move, reporting that DeepSeek's coding model sells output at a large discount to premium frontier models while closing enough of the performance gap to pressure buyers toward routing and price shopping.

Verified 12:04 AM PDT · 3 original sources

The evidence

What the reporting establishes

What happened

DeepSeek's July 31 changelog says the official V4-Flash API entered public beta, keeps the `deepseek-v4-flash` model name, adds stronger agent benchmark results, natively supports the Responses API format, and is specifically adapted for Codex-style workflows. DeepSeek's Responses API guide says V4-Flash is currently the supported model for that format, with V4-Pro support expected in early August. Axios framed the release as a price-war move, reporting that DeepSeek's coding model sells output at a large discount to premium frontier models while closing enough of the performance gap to pressure buyers toward routing and price shopping.

Pressure point

DeepSeek's benchmark table is a vendor claim, and some listed tests are internal or depend on unreleased harness details. The pricing pressure is still real if buyers can swap models behind a common protocol, but security, provenance, jurisdiction, availability, and support quality determine whether cheap agent inference is actually substitutable.

What to watch

Independent benchmark replications, V4-Pro's promised Responses API support, failure rates in long-running coding-agent tasks, buyer adoption through routers, and whether U.S. labs answer with durable price cuts or tighter differentiation on safety and enterprise controls.

Audit the story

Original sources

Company claims remain company claims. Follow the reporting and judge the evidence directly.

  1. DeepSeek API DocsChange Log
  2. DeepSeek API DocsUsing the Responses API
  3. AxiosDeepSeek's new bargain model accelerates AI's race to zero

Continue the morning

Five stories. One sourced briefing.

Read the full editionListen to the daily audio →