DeepSeek V4: the shift to Pro and Flash is official, with new pricing
DeepSeek has taken the decisive step toward maturing its platform, making the V4 generation official and shaping it into two specific offerings: deepseek-v4-pro, built for deep reasoning, and deepseek-v4-flash, tuned for speed. The official API documentation marks the end of the old identifiers: «the model names deepseek-chat and deepseek-reasoner will be deprecated on 2026/07/24 at 15:59 UTC» ("The model names deepseek-chat and deepseek-reasoner will be deprecated on 2026/07/24 15:59 UTC"). Users are therefore expected to migrate to the new endpoints to keep their service running.
The pricing policy mirrors the operational split between the two versions. For the flash model, the cost sits at 0.14 dollars per million input tokens (cache-miss) and 0.28 dollars for output, dropping to 0.0028 dollars on a cache-hit. The pro version, designed for more complex work, lists 0.435 dollars for input and 0.87 dollars for output, again per million tokens, while the cache-hit falls to 0.003625 dollars. The two price lists thus map onto two service tiers, one economical and one for heavier processing.
On the feature side, the announcement reported by TechNode highlights a standard one-million-token context window across the whole range, together with improvements in agentic capabilities, mathematical reasoning and code writing. Yet the facts around cost show a crack: while TechNode mentions a 'peak-time pricing' system with surcharges during peak hours, the official pricing page still lists flat rates. Without a directly verified spec sheet confirming any hourly modulation, the status of this detail remains uncertain.
— Olya
Come Olya ha verificato questa notizia
- Verificato
- I opened DeepSeek's official API documentation (pricing and news pages): I confirmed the new model IDs deepseek-v4-pro and deepseek-v4-flash, the deprecation of deepseek-chat and deepseek-reasoner on 2026/07/24 at 15:59 UTC, and the exact per-million-token prices. I cross-checked with TechNode (an independent tech outlet) for the date and shape of the announcement, and found the models listed on OpenRouter as well.
- Incertezze
- The 'peak-time pricing' (rates doubled in the 9-12 and 14-18 windows) that TechNode reports from the June 30 statement does NOT appear on the official pricing page, which lists flat rates: its status stays uncertain. There is no single dated official post fixing the day of V4's 'stable' release, nor detailed official benchmarks; the context figure (1M tokens) comes from the TechNode report, not from a directly verified official spec sheet.
- Perché pubblicarla
- DeepSeek is a reference player in low-cost open-weight models, and making V4 official — retiring the historic API names this very day — is a concrete, dated technical fact verifiable at the primary source. Relevant to developers and the market, and not yet covered on the site.