DeepSeek-V4-Pro GA Release
If your business uses or is evaluating DeepSeek's API for coding or agent tools, this model is now generally available with a dial for how hard it "thinks" per task, plus a new pricing structure from 16 August that's up to 50% cheaper if you can shift work to off-peak hours — worth a look if you're cost-sensitive on API spend.
What changed
Source summary
DeepSeek has made DeepSeek-V4-Pro generally available, adding adjustable "reasoning effort" levels (low, high, max) to tune performance against task complexity, plus native support for OpenAI's Responses API for easier integration with tools like Codex. New pricing takes effect 16 August 2026, introducing peak and off-peak tiers with off-peak pricing 50% below peak rates.
Why it matters to your work
Practical interpretation
If your business uses or is evaluating DeepSeek's API for coding or agent tools, this model is now generally available with a dial for how hard it "thinks" per task, plus a new pricing structure from 16 August that's up to 50% cheaper if you can shift work to off-peak hours — worth a look if you're cost-sensitive on API spend.
When this could help
Use low effort for simple lookups, high for everyday agent workflows, and max when a task is genuinely complex and worth the extra latency/cost. Switch to the Responses API path specifically when wiring DeepSeek into Codex or other Responses-API-based agent tooling.
DeepSeek-V4-Pro is now generally available with adjustable reasoning effort (low/high/max) so you can trade speed for depth per task, plus native OpenAI Responses API support with one-click Codex setup — removing custom adapter work for teams already on the Responses API pattern.
How to get started
Select Expert Mode in the DeepSeek app or web client to access V4-Pro.
Provider guidance ↗Set the reasoning effort parameter (low, high, or max) per the reasoning/thinking-mode guide to match task complexity.
Provider guidance ↗For Codex or other Responses-API-based agents, follow the Codex agent integration quick-start for one-click setup against DeepSeek's native Responses API support.
Provider guidance ↗
Build it into your routine
SCALA suggested workflow
Routing by task complexity
Default agent calls to 'low' reasoning effort for simple lookups and bump to 'high' or 'max' only when a task is flagged complex, to control latency and cost.
What to check
- New peak/off-peak pricing structure begins 16 August 2026 — off-peak workloads must be scheduled deliberately to capture the 50% discount. Source ↗
Sources & provider guides
Official documentation checked 12 Oct 2026. SCALA prompts and workflows are suggestions, not claims of hands-on testing.
Source & availability
This is a summary of a published provider announcement, with practical context from SCALA’s feed.
Plan, region, workspace and rollout conditions may affect access in Australia. Check the original source for current availability.
api-docs.deepseek.comThis update is primarily for technical teams.