AI Model Release Tracker

DeepSeek-V4-ProOpen Weight

Released

DeepSeek-V4-Pro is an AI model released by DeepSeek on Friday, Apr 24 2026, 144 days after DeepSeek-V3.2. It is an open-weight model — the trained weights are available to download and run. It is a 1.6T parameterThe learned values that shape how a model responds. B means billion and T means trillion. For mixture-of-experts models, the total count is larger than the number active for each token. Parameter count alone does not measure quality. model with a 1M token context windowHow much information the model can work with in one request, measured in tokens — pieces of words or other data. This includes your question, attachments, conversation and reply. Limits can vary by model variant and provider.. Benchmark results (shown below) cover LiveCodeBench, CheatBench, BullshitBench v2, BrowseComp, and GPQA Diamond.

Available from

Relace—$0.90$1.501Mfp4
StreamLake—$0.9553$1.91051Mfp8
GMICloud—$0.957$1.9141Mfp8
DigitalOcean—$1.044$2.0881M—
Reka—$1.05$2.101M—
Cloudflare—$1.15$2.551M—
DeepInfra—$1.30$2.601Mfp8
Alibaba—$1.416$2.8321Mfp8
Alibabaus-east-1$1.56$3.121Mfp8
SiliconFlow—$1.5016$3.1351Mfp8
Novita—$1.60$3.201Mfp8
Venice—$1.65$3.3011M—
AtlasCloud—$1.68$3.381Mfp4
Baidu—$1.69$3.381Mfp8
Azureus$1.91$3.831M—
USD per 1M tokensEvery price here is for one million tokens. A token is roughly three-quarters of a word, so a million tokens is about 750,000 words of text.Checked October 21, 2026Third-party rates from openrouter.ai, not the lab’s own list price.

Benchmarks

Coding

LiveCodeBenchCompetitive coding — Coding problems published so recently the AI can't have seen them in training — a contamination-free test of raw programming skill. Higher is better.via BenchLM
93.5%
#1

Agentic & tool use

BrowseCompWeb browsing — Can the AI browse the web and track down hard-to-find answers? Higher is better.via BenchLM
83.4%
#15 of 31

Reasoning & science

GPQA DiamondScience — Graduate-level science questions in biology, physics, and chemistry — hard enough that subject-matter PhDs score around 65%. Higher is better.
90.1%
#17 of 66

Robustness

CheatBenchCheating rate — Measures how often AI agents try to cheat on difficult assignments, such as reading hidden answers, copying work or manipulating grading. The overall score gives equal weight to ten categories; the sycophancy category measures how far an agent shifts its beliefs toward a user's stated views. Scores describe each model in its tested agent setup, and task success is measured separately. Lower is better.via CheatBench
75.1%
#9 of 13
BullshitBench v2Nonsense detection — Given a confidently-worded but nonsensical prompt, does the AI spot that it makes no sense and push back — instead of playing along and inventing an answer? The score is how often it clearly called out the nonsense. Higher is better.
14%
#78 of 97

Source: CheatBench, retrieved 25 September 2026 · Source: BenchLM, retrieved 11 July 2026. Other sources are identified on the linked benchmark pages.

Compare DeepSeek-V4-Pro with

DeepSeek-V4-Pro

Suggested comparisons

DeepSeek-V4-ProvsDeepSeek-V4.1-Flash-1017DeepSeek-V4-ProvsGPT-6.1 SolDeepSeek-V4-ProvsClaude Haiku 5.5DeepSeek-V4-ProvsGemini 4 ArgonDeepSeek-V4-ProvsMuse Spark OpenDeepSeek-V4-ProvsGrok 4.7DeepSeek-V4-ProvsMistral Large 4DeepSeek-V4-ProvsKimi K3DeepSeek-V4-ProvsGLM-5.3-FlashDeepSeek-V4-ProvsQwen3.8-Max-0902DeepSeek-V4-ProvsNemotron 3.5 Super VL

About

DeepSeek V4 Pro was introduced in the April 2026 V4 preview alongside V4 Flash. Pro used a mixture-of-experts architecture with 1.6 trillion total parameters and about 49 billion active per token. The preview supported a one-million-token context and offered thinking and non-thinking modes.

The V4 generation combined token-wise compression with DeepSeek Sparse Attention to reduce the cost of processing long inputs. Pro was available through DeepSeek’s hosted services and as downloadable weights. This entry represents the original preview checkpoint; the later V4-Pro-0813 release was a separate update, so its benchmarks and reasoning controls should not be assigned to the April build.

Frequently asked questions

DeepSeek-V4-Pro was released by DeepSeek on Friday, Apr 24 2026.

All DeepSeek releases

24 tracked