vs
12:3
overall capability wins (2 ties across 17 scored). DeepSeek V4 Flash 0731 leads overall.
- Reasoning4:1
- Coding3:1
- Agents3:0
- Tool use2:0
- Composite indices2:0
- Pricing0:3
- Specs0:1
Verdict
DeepSeek V4 Flash 0731 leads coding (widest gap: +41.6 on DeepSWE); DeepSeek V4 Pro is ~1.2x cheaper on a blended token basis than DeepSeek V4 Flash 0731.
- DeepSeek V4 Flash 0731 leads coding (widest gap: +41.6 on DeepSWE)
- DeepSeek V4 Pro is ~1.2x cheaper on a blended token basis than DeepSeek V4 Flash 0731
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- DeepSeek V4 Flash 0731
- $0.77
- DeepSeek V4 Pro
- $0.65
DeepSeek V4 Pro is about 1.2x cheaper on this workload.
Price per 1M tokens
Full comparison
Organization
DeepSeek
DeepSeek
Family
DeepSeek
DeepSeek
License
MIT
MIT
Open weights
Yes
Yes
Release
Jul 31, 2026
Apr 24, 2026
Knowledge cutoff
-
-
API / provider
DeepSeek
DeepSeek
Modalities
text → text
text → text
Specs
Context window
1M
1M
Max output
384K
384K
Parameters
304B
1.6T (49B act.)
Pricing
Input $/1M
$0.44
$0.43
Output $/1M
$1.32
$0.87
Blended $/1M (3∶1)
$0.66
$0.54
Speed
tok/s
-
85
TTFT (s)
-
0.4
Reasoning
MMLU-Pro
—
—
GPQA Diamond
90.8
88.8
Humanity's Last Exam
38.6
37.5
AIME 2025
—
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
40.4
43.0
AA-LCR v1.1
79.7
74.7
CritPt
16.6
12.9
MMMU-Pro
—
—
IFBench
—
76.5
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
—
—
SWE-bench Multilingual
—
—
LiveCodeBench
—
—
Terminal-Bench 2.1
78.7
64.0
Aider Polyglot
—
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
50.3
50.8
CursorBench
—
—
SWE-Rebench
—
41.4
NL2Repo-Bench
54.2
38.5
DeepSWE
54.4
12.8
WebDev Arena
—
1,446
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
—
1,457
| Metric | DeepSeek V4 Flash 0731 | DeepSeek V4 Pro | Delta |
|---|---|---|---|
| Identity | |||
| Organization | DeepSeek | DeepSeek | — |
| Family | DeepSeek | DeepSeek | — |
| License | MIT | MIT | — |
| Open weights | Yes | Yes | — |
| Release | Jul 31, 2026 | Apr 24, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | DeepSeek | DeepSeek | — |
| Modalities | text → text | text → text | — |
| Specs | |||
| Context window | 1M | 1M | tie |
| Max output | 384K | 384K | tie |
| Parameters | 304B | 1.6T (49B act.) | B +1K |
| Pricing | |||
| Input $/1M | $0.44 | $0.43 | B +$0.005 |
| Output $/1M | $1.32 | $0.87 | B +$0.45 |
| Blended $/1M (3∶1) | $0.66 | $0.54 | B +$0.12 |
| Speed | |||
| tok/s | - | 85 | — |
| TTFT (s) | - | 0.4 | — |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | 90.8 | 88.8 | A +2.0 pts |
| Humanity's Last Exam | 38.6 | 37.5 | A +1.1 pts |
| AIME 2025 | — | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 40.4 | 43.0 | B +2.6 pts |
| AA-LCR v1.1 | 79.7 | 74.7 | A +5.0 pts |
| CritPt | 16.6 | 12.9 | A +3.7 pts |
| MMMU-Pro | — | — | — |
| IFBench | — | 76.5 | — |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | — | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | — | — | — |
| Terminal-Bench 2.1 | 78.7 | 64.0 | A +14.7 pts |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | 50.3 | 50.8 | B +0.5 pts |
| CursorBench | — | — | — |
| SWE-Rebench | — | 41.4 | — |
| NL2Repo-Bench | 54.2 | 38.5 | A +15.7 pts |
| DeepSWE | 54.4 | 12.8 | A +41.6 pts |
| WebDev Arena | — | 1,446 | — |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | — | 1,457 | — |
DeepSeek V4 Flash 0731
- Context
- 1M / 384K out
- Parameters
- 304B
- Price
- $0.44 / $1.32 peak · $0.22 / $0.66 off-peak
- Speed
- —
- Modalities
- text → text
- License
- MIT
Official DeepSeek-V4-Flash release (Jul 31, 2026 public beta) superseding the preview - same base architecture plus a DSpark speculative-decoding module (304B total), MIT weights on Hugging Face, and agent benchmarks well above V4-Pro-Preview (Terminal-Bench 2.1 82.7, NL2Repo 54.2). Peak/off-peak API pricing took effect Aug 16, 2026: $0.44 in / $1.32 out peak, $0.22 in / $0.66 out off-peak (per 1M tokens).
DeepSeek V4 Pro
- Context
- 1M / 384K out
- Parameters
- 1.6T (49B act.)
- Price
- $0.43 / $0.87
- Speed
- 85 tok/s · 0.4s TTFT
- Modalities
- text → text
- License
- MIT
Open-weight frontier MoE - exceptional quality-per-dollar with MIT license and 1M context. Scores shown are the original preview; the GA 0813 release improves agentic coding substantially.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.