How long it has been since the last release against the line's usual gap and its recent ones — the rhythm, not a date.
Newest first — what each version changed versus the one before, in DeepSeek's own numbers.
DeepSeek's V4-Pro leaves preview with huge agentic gains — terminal, long-running coding and security tasks all jump — and stays free under the MIT license.
Compared with DeepSeek-V4-Pro (preview) · the preview it replaces
Finishes almost 9 in 10 multi-step terminal tasks
Terminal Bench 2.1 — running real commands to complete a task
87.9%
▲ up from 72.1% · +16 points
Five times better at long-running coding work
DeepSWE — multi-hour software-engineering tasks
62.7%
▲ up from 12.8% · about 5×
Free to download and run
License — open weights, commercial use allowed
MIT
same license as the preview
DeepSeek's new open flagship: about 8 times more text in memory than V3.2, while using a fraction of the computing power — and it solves about 8 in 10 real-world coding tasks.
Compared with DeepSeek-V3.2 · the version it replaces
Holds about 8 times more text in memory at once
Context window — how much text the model can read at one time
1M tokens
▲ up from 128K · about 8× more
Uses far less computing power to run at full memory
Compute at 1M context — the processing work needed to run it
27% of V3.2
▲ vs V3.2 · same 1M context
Costs developers about 14 cents per million input tokens
Price per 1M tokens — what developers pay for the Flash version's input
$0.14 / M
✦ launch price — nothing earlier to compare
One email a week: what AI models shipped, the actual differences vs the previous version, and what's coming next. No filler.