Official DeepSeek V4 Flash release with enhanced agentic capabilities and integrated DSpark speculative decoding
Specification
most-agreed values
Context
1M
Max output
82K
Released
2026-07-31
Knowledge cutoff
2025-05
Retires
—
Open weights
yes
Input
text
Output
text
Price
US dollars per million tokens · most-agreed
Input
$0.20
Output
$0.50
Cache read
$0.10
Quality
8 benchmarks · 2 sources
Composite
81st
percentile of 334 scored models
Rank
66 / 334
±0.08 sd
Evidence
8 × 2
3 or more benchmarks, from 2 or more sources
Effort range
—
one configuration only
reasoning93rd3/4 bench
math77th3/7 bench
knowledge35th1/1 bench
general41st1/1 bench
Every score
one row per source and configuration — nothing averaged away
| Benchmark | Score | Configuration | Source | Run |
|---|---|---|---|---|
| OTIS Mock AIME 2024-2025 | 94.4 ±2.1 | max | Epoch AI | 2026-08-02 |
| GPQA diamond | 91.0 ±1.8 | max | Epoch AI | 2026-08-02 |
| Chess Puzzles | 33.0 ±4.7 | max | Epoch AI | 2026-08-02 |
| Mystery Game Puzzles | 34.0 ±4.8 | max | Epoch AI | 2026-08-05 |
| FrontierMath-Tiers-1-3-v2 | 57.5 ±2.9 | max | Epoch AI | 2026-08-02 |
| LiveBench | 74.8 | default · 23 runs | LiveBench | 2026-06-25 |
| FrontierMath-Tier-4-v2 | 24.4 ±6.8 | max | Epoch AI | 2026-08-02 |
| SimpleQA Verified | 33.6 ±1.5 | max | Epoch AI | 2026-08-27 |
The leading figure for a benchmark is the median across its configurations, so one heroic high-effort run cannot set the number.
Available from 1 host
| Host | In $/M | Out $/M | Cache rd | Cache wr | Context | Output | Retires |
|---|---|---|---|---|---|---|---|
| aki-io | $0.20 | $0.50 | $0.10 | — | 1M | 82K | — |
Other listings of this model
same model, different key
More from DeepSeek
most-hosted first