DeepSeek Rolls V4-Pro Into General Availability Across App, Web and API
The agent-focused model keeps its existing API name while adding native Responses API support, new reasoning controls and a revised time-based pricing schedule.
DeepSeek rolled DeepSeek-V4-Pro into general availability on August 13, 2026, across its app, web service and API. The release completes a checkpoint the company left open when it updated V4-Flash in July, and it matters to developers because the existing `deepseek-v4-pro` API name now resolves to the GA version without a new calling pattern.
What changed in the V4-Pro release
The official DeepSeek API changelog says V4-Pro now supports the OpenAI Responses API format natively and has been adapted for Codex workflows. That gives agent developers a second interface option alongside DeepSeek's existing API patterns, while preserving the same model identifier for current integrations.
DeepSeek also added three thinking-effort settings for both V4-Pro and V4-Flash: low, high and max. These controls let an application choose how much reasoning effort to request for different tasks, but the changelog does not describe a fixed latency, token, quality or cost trade-off for each level. Teams therefore need to measure those settings against their own workloads rather than infer a universal best mode.
The same release notice describes a move to peak and off-peak API pricing, with off-peak rates set at half the peak-hour rates. DeepSeek scheduled that pricing change for 16:00 UTC on August 16. The changelog establishes the intended schedule, but it does not show an invoice-level audit or prove how every account and region was billed after the change.
Why alias continuity still requires testing
Keeping the `deepseek-v4-pro` name reduces migration work, but it also means an unchanged model parameter can begin returning a materially updated service. That is useful for fast adoption and risky for teams that treat a model alias as a fixed artifact. Agent loops can be sensitive to changes in tool selection, response structure, token use and retry behaviour even when headline capability improves.
Developers using V4-Pro in production should re-run task suites, structured-output checks, tool-call validation and safety tests against the served GA model. Applications that adopt the Responses API should also verify how their existing state, tools and error handling map onto DeepSeek's implementation instead of assuming perfect compatibility from the format name alone.
The time-based pricing model adds another operational consideration. Batchable work may be shifted into cheaper periods, but interactive applications may not have that flexibility. A lower off-peak rate is therefore not the same as a guaranteed lower blended bill; usage timing, reasoning effort, retries and output length still determine the result.
Evidence and limits
DeepSeek reports stronger agent performance across terminal, repository, cybersecurity, tool-use and automation evaluations. Those results come from the vendor's own changelog. This article does not treat them as independently replicated performance or as proof that the GA model will improve every coding-agent deployment.
The captured release page does not provide a full model card, architecture description, regional availability matrix, production error rate or independent safety evaluation. It also does not quantify how low, high and max effort change latency or token consumption. Those missing details matter for organizations comparing V4-Pro with another model or deciding whether the GA label is sufficient for a controlled production rollout.
Status
Confirmed. DeepSeek's official API documentation records the August 13 GA rollout, the unchanged model name, native Responses API support, reasoning-effort controls and the pricing schedule. Internal confidence is medium because the capability and rollout evidence is first-party and has not been independently corroborated here.
Sources
Update note: Last reviewed 2026-08-17. We will revise this post if DeepSeek publishes a model card, compatibility details, independent evaluations or a changed availability statement.
Sources
- DeepSeek API change log — official
Drafted with AI assistance from source briefs; reviewed for citation completeness and label accuracy.