vs
0:9
overall capability wins (0 ties across 9 scored). DeepSeek V4 Pro leads overall.
- Reasoning0:5
- Coding0:1
- Pricing1:2
- Speed0:2
- Specs0:3
Verdict
DeepSeek V4 Pro leads coding (widest gap: +19.5 on SciCode); DeepSeek V4 Pro offers ~7.8x larger context (1M vs 128K); DeepSeek V4 Pro allows ~46.9x more max output (384K vs 8K).
- DeepSeek V4 Pro leads coding (widest gap: +19.5 on SciCode)
- DeepSeek V4 Pro offers ~7.8x larger context (1M vs 128K)
- DeepSeek V4 Pro allows ~46.9x more max output (384K vs 8K)
- DeepSeek V4 Pro is ~1.5x cheaper on a blended token basis than DeepSeek-R1-Distill-Llama-70B
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- DeepSeek-R1-Distill-Llama-70B
- $1
- DeepSeek V4 Pro
- $0.65
DeepSeek V4 Pro is about 1.5x cheaper on this workload.
Price per 1M tokens
Full comparison
Organization
DeepSeek
DeepSeek
Family
DeepSeek
DeepSeek
License
MIT
MIT
Open weights
Yes
Yes
Release
Jan 20, 2025
Apr 24, 2026
Knowledge cutoff
-
-
API / provider
Together / Fireworks (ref.)
DeepSeek
Modalities
text → text
text → text
Specs
Context window
128K
1M
Max output
8K
384K
Parameters
70B
1.6T (49B act.)
Pricing
Input $/1M
$0.80
$0.43
Output $/1M
$0.80
$0.87
Blended $/1M (3∶1)
$0.80
$0.54
Speed
tok/s
55
85
TTFT (s)
0.45
0.4
Reasoning
MMLU-Pro
79.5
—
GPQA Diamond
40.2
88.8
Humanity's Last Exam
5.1
37.5
AIME 2025
53.7
—
MATH-500
94.5
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
19.2
43.0
AA-LCR v1.1
10.0
74.7
CritPt
—
12.9
MMMU-Pro
—
—
IFBench
27.6
76.5
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
—
—
SWE-bench Multilingual
—
—
LiveCodeBench
57.5
—
Terminal-Bench 2.1
—
64.0
Aider Polyglot
—
—
Terminal-Bench 3
—
—
BigCodeBench
35.3
—
SciCode
31.3
50.8
CursorBench
—
—
SWE-Rebench
—
41.4
NL2Repo-Bench
—
38.5
DeepSWE
—
12.8
WebDev Arena
—
1,446
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
—
1,457
| Metric | DeepSeek-R1-Distill-Llama-70B | DeepSeek V4 Pro | Delta |
|---|---|---|---|
| Identity | |||
| Organization | DeepSeek | DeepSeek | — |
| Family | DeepSeek | DeepSeek | — |
| License | MIT | MIT | — |
| Open weights | Yes | Yes | — |
| Release | Jan 20, 2025 | Apr 24, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | Together / Fireworks (ref.) | DeepSeek | — |
| Modalities | text → text | text → text | — |
| Specs | |||
| Context window | 128K | 1M | B +872K |
| Max output | 8K | 384K | B +376K |
| Parameters | 70B | 1.6T (49B act.) | B +2K |
| Pricing | |||
| Input $/1M | $0.80 | $0.43 | B +$0.37 |
| Output $/1M | $0.80 | $0.87 | A +$0.07 |
| Blended $/1M (3∶1) | $0.80 | $0.54 | B +$0.26 |
| Speed | |||
| tok/s | 55 | 85 | B +30 |
| TTFT (s) | 0.45 | 0.4 | B +0.05 |
| Reasoning | |||
| MMLU-Pro | 79.5 | — | — |
| GPQA Diamond | 40.2 | 88.8 | B +48.6 pts |
| Humanity's Last Exam | 5.1 | 37.5 | B +32.4 pts |
| AIME 2025 | 53.7 | — | — |
| MATH-500 | 94.5 | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 19.2 | 43.0 | B +23.8 pts |
| AA-LCR v1.1 | 10.0 | 74.7 | B +64.7 pts |
| CritPt | — | 12.9 | — |
| MMMU-Pro | — | — | — |
| IFBench | 27.6 | 76.5 | B +48.9 pts |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | — | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | 57.5 | — | — |
| Terminal-Bench 2.1 | — | 64.0 | — |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | 35.3 | — | — |
| SciCode | 31.3 | 50.8 | B +19.5 pts |
| CursorBench | — | — | — |
| SWE-Rebench | — | 41.4 | — |
| NL2Repo-Bench | — | 38.5 | — |
| DeepSWE | — | 12.8 | — |
| WebDev Arena | — | 1,446 | — |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | — | 1,457 | — |
DeepSeek-R1-Distill-Llama-70B
- Context
- 128K / 8K out
- Parameters
- 70B
- Price
- $0.80 / $0.80
- Speed
- 55 tok/s · 0.45s TTFT
- Modalities
- text → text
- License
- MIT
Llama-70B distilled from R1 reasoning traces — open reasoning without full R1 compute cost.
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.