DeepSeek

DeepSeek · flagship line · open weights

17 releases since 2023Latest: DeepSeek-V4-Pro-0813

The newest DeepSeek model is DeepSeek-V4-Pro-0813, released on August 13, 2026. It has been 72 days since then, longer than the line's usual gap of 63 days. DeepSeek has officially announced DeepSeek-V4.1-Pro; on Sep 10, 2026 it said: “all deepseek-v4-pro requests will route to V4.1-Flash at V4.1-Flash rates. This will continue until V4.1-Pro launches”.

Where the line stands

How long it has been since the last release against the line's usual gap and its recent ones — the rhythm, not a date.

DeepSeek

DeepSeek · flagship line · open weights

Latest: DeepSeek-V4-Pro-0813 · Aug 13, 2026

AnnouncedDeepSeek-V4.1-Pro · announced Sep 10, 2026, 44 days ago

“all deepseek-v4-pro requests will route to V4.1-Flash at V4.1-Flash rates. This will continue until V4.1-Pro launches.” — DeepSeek, Sep 10, 2026 · DeepSeek API docs

The general-availability release of V4-Pro scores 87.9% on Terminal Bench 2.1, up from 72.1% for the April preview in DeepSeek's tests, and keeps the MIT license.

What changed in DeepSeek-V4-Pro-0813 →

Days since last release

days

Usual gap

days

Recent gaps

DeepSeek-V4-Pro-0813 · Aug 2026

111

DeepSeek-V4 · Apr 2026

144

DeepSeek-V3.2-Speciale · Dec 2025

DeepSeek-V3.2-Exp · Sep 2025

Get new AI models by email

Every Monday, only when something shipped: the AI models that came out, what each one changed, and which lines are due next.

We email you a link to confirm first · unsubscribe anytime · read past issues · about the newsletter · RSS for the DeepSeek line

Every DeepSeek release

Newest first — what each version changed versus the one before, in DeepSeek's own numbers.

Aug

2026

DeepSeek-V4-Pro-0813 DeepSeek · DeepSeekOpen weights

The general-availability release of V4-Pro scores 87.9% on Terminal Bench 2.1, up from 72.1% for the April preview in DeepSeek's tests, and keeps the MIT license.

Compared with DeepSeek-V4 · 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

Scores about five times higher on long-running coding work

DeepSWE — multi-hour software-engineering tasks

62.7%

▲ up from 12.8% · about 5 times

Costs developers at least three times more than the preview at peak hours

Price per 1M tokens — what developers pay, input / output, at peak hours

$1.32 / $3.96

up from $0.435 / $0.87 · from Aug 16 · half price off-peak

Sources: DeepSeek · Hugging Face

Apr

2026

DeepSeek-V4 DeepSeek · DeepSeekOpen weights

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 times more

Needs about a quarter of V3.2's computing power at full memory

Compute at 1M context — the processing work needed to run it

27% of V3.2

▲ vs V3.2 · same 1M context

Listed at $1.74 per million input tokens at launch, then cut by 75% a day later

Price per 1M tokens — what developers pay for V4-Pro, input / output

$1.74 / $3.48

✦ launch list price · 75% off from Apr 25, made permanent in May at $0.435 / $0.87

Sources: DeepSeek · Hugging Face · Engadget

Dec

2025

DeepSeek-V3.2 DeepSeek · DeepSeekOpen weights

Dec

2025

DeepSeek-V3.2-Speciale DeepSeek · DeepSeekOpen weights

Sep

2025

DeepSeek-V3.2-Exp DeepSeek · DeepSeekOpen weights

Sep

2025

DeepSeek-V3.1-Terminus DeepSeek · DeepSeekOpen weights

Aug

2025

DeepSeek-V3.1 DeepSeek · DeepSeekOpen weights

May

2025

DeepSeek-R1-0528 DeepSeek · DeepSeekOpen weights

Mar

2025

DeepSeek-V3-0324 DeepSeek · DeepSeekOpen weights

Jan

2025

DeepSeek-R1 DeepSeek · DeepSeekOpen weights

DeepSeek's reasoning model, trained from the DeepSeek-V3 base with large-scale reinforcement learning and released under the MIT license, roughly doubles V3's AIME 2024 math score at about twice V3's API price.

Compared with DeepSeek-V3 · the model it is built on

Solves 8 in 10 problems from a hard US high-school math contest

AIME 2024 — hard problems from a US high-school math contest

79.8%

▲ up from 39.2% for DeepSeek-V3 · +40.6 points

Costs developers about twice as much as DeepSeek-V3

API price per 1M tokens — input (cache miss) / output

$0.55 / $2.19

about double V3's $0.27 / $1.10, its list price from Feb 8, 2025

Sources: DeepSeek · Hugging Face · DeepSeek

Dec

2024