0

DeepSeek V4 Pro 0423

DeepSeek's April 2026 next-gen open-weights flagship - 1.6T-total / 49B-active MoE with 1M context and DeepSeek Sparse Attention.

Open
Publisher
DeepSeek
Family
Deepseek
Params
1.6T total / 49B active
Context
1.0M
Input
Output
Providers
deepseektogetherfireworksself-host
License
mit
Released
13 Aug 2026
Canonical
api-docs.deepseek.com/news/news260424

Cite

Notes

Only stored in your browser.

Attribution

Benchmark scores
AALiveBenchVaultmatharenaVals AIEQ-BenchLech Mazur
Attribution policy →
Context
1.0M
Weights
Open
downloadable
Input
$1.32
/1M tok
Output
$3.96
/1M tok
Blended
$1.98
#421/ 499/1M · 3:1
Speed
96
#57/ 498tok/s
Cache read
$0.03
0.02x input

Price & speed via Artificial Analysis · ranked across tracked models · caching via OpenRouter

Reasoning effortIndexPriceSpeed
max36.0$1.9896 t/s
high30.1$0.540 t/s
Non-reasoning20.8$0.540 t/s

Intelligence Index per reasoning-effort setting, via Artificial Analysis

Reported on 27 evals across 12 domains - best 96.2% on τ²-bench (Tau²-bench) · #13 of 321

Reported eval scores

27
SWE-bench80.6
GPQA Diamond92.8
SciCode51
τ²-bench (Tau²-bench)96.2
IFBench76.5
Terminal-Bench (Hard)46.2
LiveBench - Language78.1
LiveBench - Math90.7
LiveBench - Data Analysis74.5
LiveBench - Reasoning82.7
LiveBench - Coding70
LiveBench - Instruction Following62.4
Humanity's Last Exam (HLE)41
LiveBench76.4
Vals Index38.6
EQ-Bench Longform Creative Writing75.6
EQ-Bench Creative Writing v31553.2
Creative Story-Writing Benchmark42
MathArena76.5
CorpFin v261.4