DeepSeek V4 Arrives Mid-July — and the July 24 API Deadline Is Real

DeepSeek V4 is rolling out this week, and with it comes a deadline that developers using the old API names need to take seriously. On July 24 at 15:59 UTC, the identifiers deepseek-chat and deepseek-reasoner stop working. Calls using those names will fail outright.

Two models, very different price points

V4 ships as two variants. V4-Pro is the heavyweight: 1.6 trillion total parameters with 49 billion activated per token, using a Mixture of Experts architecture that keeps actual compute costs manageable despite the parameter count. V4-Flash covers high-volume and latency-sensitive work at 284 billion total parameters, with 13 billion activated.

Pricing is the part that makes western developers pay attention. V4-Pro runs $1.74 per million input tokens and $3.48 per million output tokens. V4-Flash is $0.14 / $0.28. By comparison, GPT-5.6 Sol (OpenAI’s top reasoning tier) is priced at $5 / $30 per million tokens. V4-Flash is nearly 100x cheaper than Sol per output token.

Both models ship under an MIT license, so self-hosting is an option for teams with data residency requirements.

Context window jumps to 1M

Both V4 variants set 1M tokens as the default context floor. The previous V3.2 models topped out at 128K. For codebases or long-running workflows that push context limits, that’s a meaningful jump.

Efficiency also improved substantially. At maximum context, V4-Pro requires roughly 27% of the single-token inference compute and 10% of the KV cache that V3.2 used.

The migration is straightforward but not optional

Right now, deepseek-chat and deepseek-reasoner both route to V4-Flash as a compatibility bridge. After July 24, that bridge disappears.

The migration itself is simple: update your model parameter from deepseek-chat to deepseek-v4-flash, or to deepseek-v4-pro if you were using the heavy reasoning variant. The API endpoints don’t change — only the model identifier.

If your codebase calls DeepSeek anywhere, search for those legacy strings now. The blast radius can be larger than it looks: direct API calls, wrapper libraries, LangChain configs, and hosted tools that pass through to DeepSeek all need updating.

Peak-time pricing is new

DeepSeek also introduced differentiated pricing for the first time. Peak hours (9:00 to 12:00 and 14:00 to 18:00 daily) run at double the off-peak rate. If your workloads are flexible, scheduling jobs outside those windows cuts costs in half.

Coding performance

On Codeforces, V4-Pro scores a rating of 3,206, which works out to roughly rank 23 among human competitors. On SWE-bench Verified, the open-weights version hits 80.6%, tying with Gemini 3.1 Pro as the highest open-weights entry.

The practical implication for coding tools: V4-Flash at $0.14/$0.28 gives teams a serious code-capable model at near-commodity pricing, with V4-Pro available when harder problems need it.

Sources: DeepSeek API docs · Verdent Guides · TechNode