vs
1:9
overall capability wins (0 ties across 10 scored). DeepSeek V4 Flash 0731 leads overall.
- Reasoning0:5
- Coding0:2
- Pricing3:0
- Specs1:2
Verdict
DeepSeek V4 Flash 0731 leads coding (widest gap: +31.9 on TermBench); DeepSeek V4 Flash 0731 offers ~6.1x larger context (1M vs 164K); DeepSeek V4 Flash 0731 allows ~6.0x more max output (384K vs 64K).
- DeepSeek V4 Flash 0731 leads coding (widest gap: +31.9 on TermBench)
- DeepSeek V4 Flash 0731 offers ~6.1x larger context (1M vs 164K)
- DeepSeek V4 Flash 0731 allows ~6.0x more max output (384K vs 64K)
- DeepSeek V3.2 is ~2.1x 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 V3.2
- $0.39
- DeepSeek V4 Flash 0731
- $0.77
DeepSeek V3.2 is about 2.0x cheaper on this workload.
Price per 1M tokens
Full comparison
Organization
DeepSeek
DeepSeek
Family
DeepSeek
DeepSeek
License
MIT
MIT
Open weights
Yes
Yes
Release
Nov 20, 2025
Jul 31, 2026
Knowledge cutoff
-
-
API / provider
DeepSeek
DeepSeek
Modalities
text → text
text → text
Specs
Context window
164K
1M
Max output
64K
384K
Parameters
685B (37B act.)
304B
Pricing
Input $/1M
$0.28
$0.44
Output $/1M
$0.42
$1.32
Blended $/1M (3∶1)
$0.32
$0.66
Speed
tok/s
90
-
TTFT (s)
0.35
-
Reasoning
MMLU-Pro
86.2
—
GPQA Diamond
84.0
90.8
Humanity's Last Exam
24.6
38.6
AIME 2025
92.0
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
33.0
40.4
AA-LCR v1.1
73.3
79.7
CritPt
2.9
16.6
MMMU-Pro
—
—
IFBench
60.7
—
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
70.0
—
SWE-bench Pro
15.6
—
SWE-bench Multilingual
59.0
—
LiveCodeBench
86.2
—
Terminal-Bench 2.1
46.8
78.7
Aider Polyglot
—
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
38.9
50.3
CursorBench
—
—
SWE-Rebench
48.9
—
NL2Repo-Bench
—
54.2
DeepSWE
—
54.4
WebDev Arena
1,325
—
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
1,425
—
| Metric | DeepSeek V3.2 | DeepSeek V4 Flash 0731 | Delta |
|---|---|---|---|
| Identity | |||
| Organization | DeepSeek | DeepSeek | — |
| Family | DeepSeek | DeepSeek | — |
| License | MIT | MIT | — |
| Open weights | Yes | Yes | — |
| Release | Nov 20, 2025 | Jul 31, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | DeepSeek | DeepSeek | — |
| Modalities | text → text | text → text | — |
| Specs | |||
| Context window | 164K | 1M | B +836K |
| Max output | 64K | 384K | B +320K |
| Parameters | 685B (37B act.) | 304B | A +381 |
| Pricing | |||
| Input $/1M | $0.28 | $0.44 | A +$0.16 |
| Output $/1M | $0.42 | $1.32 | A +$0.90 |
| Blended $/1M (3∶1) | $0.32 | $0.66 | A +$0.35 |
| Speed | |||
| tok/s | 90 | - | — |
| TTFT (s) | 0.35 | - | — |
| Reasoning | |||
| MMLU-Pro | 86.2 | — | — |
| GPQA Diamond | 84.0 | 90.8 | B +6.8 pts |
| Humanity's Last Exam | 24.6 | 38.6 | B +14.0 pts |
| AIME 2025 | 92.0 | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 33.0 | 40.4 | B +7.4 pts |
| AA-LCR v1.1 | 73.3 | 79.7 | B +6.4 pts |
| CritPt | 2.9 | 16.6 | B +13.7 pts |
| MMMU-Pro | — | — | — |
| IFBench | 60.7 | — | — |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | 70.0 | — | — |
| SWE-bench Pro | 15.6 | — | — |
| SWE-bench Multilingual | 59.0 | — | — |
| LiveCodeBench | 86.2 | — | — |
| Terminal-Bench 2.1 | 46.8 | 78.7 | B +31.9 pts |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | 38.9 | 50.3 | B +11.4 pts |
| CursorBench | — | — | — |
| SWE-Rebench | 48.9 | — | — |
| NL2Repo-Bench | — | 54.2 | — |
| DeepSWE | — | 54.4 | — |
| WebDev Arena | 1,325 | — | — |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | 1,425 | — | — |
DeepSeek V3.2
- Context
- 164K / 64K out
- Parameters
- 685B (37B act.)
- Price
- $0.28 / $0.42
- Speed
- 90 tok/s · 0.35s TTFT
- Modalities
- text → text
- License
- MIT
Prior DeepSeek V3.x generation - still excellent open-weight coding value.
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.