DeepSeek · open weights
DeepSeek V4-Pro
DeepSeek's larger V4 sibling: MIT open weights, text-only, 1M context, and 384K max output. Hosted billing is peak/off-peak after 2026-08-16 rather than one standard-tier list rate.
·MIT·1.6T total / 49B active (MoE)
text
- Practical fit
- The self-host and hosted path when V4-Flash is not enough for harder reasoning or coding. Pin the 2026-08-13 GA checkpoint — the April preview reused the same API id.
- Deployment note
- Catalogued as 1.6T total / 49B active MoE. Hosted API id stays `deepseek-v4-pro`. Official peak/off-peak cache-miss/output rates in the catalog note: off-peak $0.66/$1.98, peak $1.32/$3.96.