vs
1:8
overall capability wins (0 ties across 9 scored). DeepSeek V4 Flash 0731 leads overall.
- Reasoning0:5
- Coding0:1
- Pricing3:0
- Specs1:2
Verdict
DeepSeek V4 Flash 0731 leads coding (widest gap: +11.2 on SciCode); DeepSeek V4 Flash 0731 offers ~6.1x larger context (1M vs 164K); DeepSeek V4 Flash 0731 allows ~11.7x more max output (384K vs 33K).
- DeepSeek V4 Flash 0731 leads coding (widest gap: +11.2 on SciCode)
- DeepSeek V4 Flash 0731 offers ~6.1x larger context (1M vs 164K)
- DeepSeek V4 Flash 0731 allows ~11.7x more max output (384K vs 33K)
- DeepSeek V3.1 is ~1.6x 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 V3.1
- $0.49
- DeepSeek V4 Flash 0731
- $0.77
DeepSeek V3.1 is about 1.6x cheaper on this workload.
Price per 1M tokens
Full comparison
Organization
DeepSeek
DeepSeek
Family
DeepSeek
DeepSeek
License
MIT
MIT
Open weights
Yes
Yes
Release
Aug 21, 2025
Jul 31, 2026
Knowledge cutoff
2025-03
-
API / provider
DeepSeek / OpenRouter (ref.)
DeepSeek
Modalities
text → text
text → text
Specs
Context window
164K
1M
Max output
33K
384K
Parameters
671B (37B act.)
304B
Pricing
Input $/1M
$0.25
$0.44
Output $/1M
$0.95
$1.32
Blended $/1M (3∶1)
$0.42
$0.66
Speed
tok/s
85
-
TTFT (s)
0.35
-
Reasoning
MMLU-Pro
84.8
—
GPQA Diamond
77.9
90.8
Humanity's Last Exam
14.3
38.6
AIME 2025
88.4
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
29.0
40.4
AA-LCR v1.1
56.7
79.7
CritPt
2.0
16.6
MMMU-Pro
—
—
IFBench
41.5
—
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
—
—
SWE-bench Multilingual
—
—
LiveCodeBench
74.8
—
Terminal-Bench 2.1
—
78.7
Aider Polyglot
76.3
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
39.1
50.3
CursorBench
—
—
SWE-Rebench
28.0
—
NL2Repo-Bench
—
54.2
DeepSWE
—
54.4
WebDev Arena
—
—
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
1,417
—
| Metric | DeepSeek V3.1 | DeepSeek V4 Flash 0731 | Delta |
|---|---|---|---|
| Identity | |||
| Organization | DeepSeek | DeepSeek | — |
| Family | DeepSeek | DeepSeek | — |
| License | MIT | MIT | — |
| Open weights | Yes | Yes | — |
| Release | Aug 21, 2025 | Jul 31, 2026 | — |
| Knowledge cutoff | 2025-03 | - | — |
| API / provider | DeepSeek / OpenRouter (ref.) | DeepSeek | — |
| Modalities | text → text | text → text | — |
| Specs | |||
| Context window | 164K | 1M | B +836K |
| Max output | 33K | 384K | B +351K |
| Parameters | 671B (37B act.) | 304B | A +367 |
| Pricing | |||
| Input $/1M | $0.25 | $0.44 | A +$0.19 |
| Output $/1M | $0.95 | $1.32 | A +$0.37 |
| Blended $/1M (3∶1) | $0.42 | $0.66 | A +$0.24 |
| Speed | |||
| tok/s | 85 | - | — |
| TTFT (s) | 0.35 | - | — |
| Reasoning | |||
| MMLU-Pro | 84.8 | — | — |
| GPQA Diamond | 77.9 | 90.8 | B +12.9 pts |
| Humanity's Last Exam | 14.3 | 38.6 | B +24.3 pts |
| AIME 2025 | 88.4 | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 29.0 | 40.4 | B +11.4 pts |
| AA-LCR v1.1 | 56.7 | 79.7 | B +23.0 pts |
| CritPt | 2.0 | 16.6 | B +14.6 pts |
| MMMU-Pro | — | — | — |
| IFBench | 41.5 | — | — |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | — | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | 74.8 | — | — |
| Terminal-Bench 2.1 | — | 78.7 | — |
| Aider Polyglot | 76.3 | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | 39.1 | 50.3 | B +11.2 pts |
| CursorBench | — | — | — |
| SWE-Rebench | 28.0 | — | — |
| NL2Repo-Bench | — | 54.2 | — |
| DeepSWE | — | 54.4 | — |
| WebDev Arena | — | — | — |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | 1,417 | — | — |
DeepSeek V3.1
- Context
- 164K / 33K out
- Parameters
- 671B (37B act.)
- Price
- $0.25 / $0.95
- Speed
- 85 tok/s · 0.35s TTFT
- Modalities
- text → text
- License
- MIT
Updated DeepSeek V3 mid-2025 release with stronger coding before V3.2 / V4; widely hosted via OpenRouter and Bedrock.
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).
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.