DeepSeek’s V4 model family moved from general availability to a hard cutover this week: the company’s legacy deepseek-chat and deepseek-reasoner model identifiers were formally retired at 15:59 UTC on July 24, 2026, meaning any application still calling those names will stop working rather than falling back to an older model.
V4 ships in two variants — V4 Pro, a 1.6-trillion-parameter mixture-of-experts model with 49 billion active parameters, and V4 Flash, a smaller 284-billion-parameter version — both offering a 1-million-token context window under an MIT license.
Since its move to general availability on July 19, early testers have reported overall performance approaching Anthropic’s Opus 4.8, with coding ability closing in on GPT-5.6 Sol, according to one detailed launch analysis, while undercutting Anthropic’s flagship Fable 5 pricing by more than 50x per token.
The number that matters most: 80.6%
V4 Pro’s 80.6% score on SWE-bench Verified, a widely used coding benchmark, ties it with Google’s Gemini 3.1 Pro for the top spot among all models — open or closed — on that particular test. Combined with pricing near $0.435 per million input tokens, V4 Pro is positioned as a genuine substitute for premium closed models on coding workloads, not merely a cheaper, weaker alternative.
The caveats worth knowing before switching
V4 ships with built-in content moderation that refuses roughly 45% of tested code-generation prompts touching sensitive content, according to independent testing — a meaningfully higher refusal rate than Western frontier models on comparable prompts, and a detail that matters for teams building tools that touch security research, exploit development, or other borderline categories.
DeepSeek is also reportedly raising its first external funding round, with Tencent and Alibaba said to be contributing a combined $1.8 billion — a shift from the company’s previously self-funded posture that may signal commercial ambitions beyond simply publishing open weights.
The retirement of the legacy endpoint names is a smaller story than the pricing shock, but it’s the one most likely to break something in production this week. Any application that hard-coded deepseek-chat or deepseek-reasoner rather than the current model identifiers will now fail outright rather than silently degrade, since DeepSeek has not left a compatibility shim in place.
Teams that migrated ahead of the July 24 deadline report the swap is mechanically simple — the base URL and request format are unchanged, only the model field needs updating — but it does require noticing the deprecation in the first place.
What to watch next
- Whether any application still hard-coded to the retired deepseek-chat or deepseek-reasoner names has already broken in production.
- Whether DeepSeek’s funding round closes and what it signals about the company’s next model cadence.
- How V4’s aggressive pricing affects renewal conversations for teams currently on Claude, GPT, or Gemini API contracts.
Helpful Sources
- DeepSeek V4 GA Is Here: Near-Opus Performance at 1/57th the Price of Fable 5 — Wan 2.7
- DeepSeek V4 (2026) — 1T Params, Benchmarks & Pricing
- DeepSeek V4 Review 2026: Benchmarks, Pricing, Architecture — AI Stack Choice
Disclaimer: This content is meant to inform and should not be considered financial advice. The views expressed in this article may include the author’s personal opinions and do not represent Times Tabloid’s opinion. Readers are advised to conduct thorough research before making any investment decisions. Any action taken by the reader is strictly at their own risk. Times Tabloid is not responsible for any financial losses.
Follow us on X, Facebook, Telegram, and Google News
