DeepSeek-V4-Pro

Larger/most-capable V4 model (~1.6T total / ~49B active params per HuggingFace model card + authoritative third-party reports; MIT License; mixed FP4/FP8). Context length 1M, max output 384K tokens. Input price $1.32/M cache-miss, $0.044/M cache-hit; output $3.96/M (USD). Supports three reasoning-effort modes (non-think / think high / think max), JSON output, tool calls; FIM completion non-thinking-mode only. Concurrency limit 500. Part of the 'DeepSeek V4 Preview' generation (released 2026-04-24), hence status=preview. Knowledge cutoff NOT officially published by DeepSeek -> left null. Prices re-verified to the cent against api-docs.deepseek.com/quick_start/pricing/ on 2026-08-09. VENDOR-DECLARED FORWARD PRICE EVENT, RESOLVED 2026-08-14 — the same footnote now carries a date and a table. Until 2026-08-13 it read only 'We plan to raise the overall pricing for DeepSeek API services in the near future, with a significant increase expected... subject to official notice', with no date and no figure. DeepSeek now states: 'DeepSeek API pricing will be updated to peak / off-peak billing, with off-peak rates at half the peak rates. Peak hours are 01:00 - 04:00 and 06:00 - 10:00 UTC (all other hours are off-peak). The new prices take effect at 16:00 UTC on August 16, 2026'. Announced rates for this model per 1M tokens: OFF-PEAK $0.022 cache-hit / $0.66 cache-miss / $1.98 out; PEAK $0.044 cache-hit / $1.32 cache-miss / $3.96 out. EXECUTED 2026-08-16: the priced fields above were re-pointed at the PEAK column on the cutover day, because DeepSeek presents peak as the rate and off-peak as half of it, not the other way round. The OFF-PEAK figure is exactly half on every axis and is what you actually pay for 17 of every 24 hours, since peak is only 01:00-04:00 and 06:00-10:00 UTC. Re-read first-hand from api-docs.deepseek.com/quick_start/pricing/ on 2026-08-16, when the page still stated the cutover as forthcoming: the fields were moved ~8 hours early, in the run of the cutover day, because this catalog is refreshed once daily at ~08:05 UTC and the alternative was to publish the superseded rates for ~16 hours after they stopped being true. THAT WINDOW IS NOW CLOSED: re-read first-hand 2026-08-17, the page presents the peak/off-peak table as the rates in force and the forward-dated announcement banner is gone; every figure above still matches it to the cent, so the fields are simply current from here on. Responses API SUPPORTED as of 2026-08-17: footnote (1) had committed to 'early August 2026' and the feature table still showed it unsupported on 2026-08-09; the table now marks Responses API and Anthropic-format API as supported for BOTH V4 models (re-read first-hand 2026-08-17). Model version string on that table: DeepSeek-V4-Pro-0813. The pricing page is served at the TRAILING-SLASH url (2026-08-11): the extensionless form returns a different document ("Your First API Call") with HTTP 200.

Source: DeepSeek official documentation ↗

Compare DeepSeek-V4-Pro head-to-head

DeepSeek-V4-Pro — questions & answers