DeepSeek’s API history is a case study in moving model aliases. On 20 January 2025, its official changelog associated deepseek-reasoner with R1. By December 2025 that same name referred to the thinking mode of DeepSeek-V3.2. An April 2026 announcement introduced V4 Pro and Flash and warned that old names would be discontinued. A September 2026 entry added V4.1 Flash. Anyone benchmarking “reasoner” across this period without a date and resolved target is mixing models.
The important dates
DeepSeek-V3.2 was published on 1 December 2025. DeepSeek-V4 Preview launched on 24 April 2026 with API identifiers deepseek-v4-pro and deepseek-v4-flash. The changelog set 24 July 2026 as the discontinuation deadline for deepseek-chat and deepseek-reasoner; the newer family appeared separately. The 10 September 2026 V4.1 Flash update added native visual understanding. Exact endpoint claims are time-sensitive and are recorded with dates rather than projected into the past.
Thinking modes
A model name and its thinking mode are separate variables. Two model settings can use different output budgets and produce different latency. When comparing reasoning models, include internal reasoning token billing where documented, emitted answer quality, tool use and end-to-end processing time. Do not report the context maximum from a page updated after the article cutoff as if it were the limit of R1 in January 2025.
Migration checklist
Replace undocumented legacy aliases with explicit supported IDs. Save example request and response metadata, audit JSON mode and tool-calling differences, and verify any OpenAI-compatible surface accepts only the parameters you send. Compatibility at the HTTP interface does not imply identical support for every option or error code.
Deployment choice
DeepSeek has published open-weight model material in addition to hosted endpoints. Review the specific checkpoint license and platform terms independently; a public API does not prove local weights are available for every minor release. Regional hosting, logging and content rules remain customer-side verification items.
What to verify before adopting this model or method
Use the exact release ID and API endpoint; confirm your geographic processing location, entitlement tier, model lifecycle, feature flags, context and output limits from the provider console. Run a reproducible evaluation on your own data and retain human oversight for consequential actions. Where a specification, price, commercial license or entitlement is not verified for your deployment, this article deliberately leaves it unasserted instead of filling a plausible number.
Documentation clarification (25 October 2026)
DeepSeek’s dated changelog states that legacy deepseek-chat/deepseek-reasoner IDs were scheduled for discontinuation on July 24, 2026; the September 10, 2026 entry says retired V4 Flash aliases temporarily route to V4.1 Flash. A compatibility redirect is not a pinned checkpoint.
Sources & Further Reading
- DeepSeek API changelog (retrieved from first-party public documentation on 2026-10-25; this is not a deployment test)
- deepseek.com — documentation/reference (supplier-provided reference; not independently checked in this pass)
- www.deepseek.com — documentation/reference (supplier-provided reference; not independently checked in this pass)
Source documentation establishes vendor-published facts, not independently measured model quality, tenant access, contractual rights or current prices. Use a dated pricing ledger and an account-specific entitlement check for deployment decisions.