vs
12:2
overall capability wins (1 ties across 15 scored). Claude Sonnet 5 leads overall.
- Reasoning4:1
- Coding5:0
- Arena1:0
- Tool use2:0
- Composite indices2:0
- Pricing0:3
- Speed1:0
- Specs0:1
Verdict
Claude Sonnet 5 leads coding (widest gap: +91.0 on WebDev Elo); Claude Sonnet 5 ranks higher on LMArena (+4 Elo); DeepSeek V4 Pro allows ~3.0x more max output (384K vs 128K).
- Claude Sonnet 5 leads coding (widest gap: +91.0 on WebDev Elo)
- Claude Sonnet 5 ranks higher on LMArena (+4 Elo)
- DeepSeek V4 Pro allows ~3.0x more max output (384K vs 128K)
- DeepSeek V4 Pro ships open weights (MIT)
- DeepSeek V4 Pro is ~7.4x cheaper on a blended token basis than Claude Sonnet 5
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Claude Sonnet 5
- $4.50
- DeepSeek V4 Pro
- $0.65
DeepSeek V4 Pro is about 6.9x cheaper on this workload.
Price per 1M tokens
Full comparison
Organization
Anthropic
DeepSeek
Family
Claude Sonnet / Haiku
DeepSeek
License
Proprietary
MIT
Open weights
No
Yes
Release
Jun 30, 2026
Apr 24, 2026
Knowledge cutoff
-
-
API / provider
Anthropic
DeepSeek
Modalities
text, image → text
text → text
Specs
Context window
1M
1M
Max output
128K
384K
Parameters
—
1.6T (49B act.)
Pricing
Input $/1M
$2
$0.43
Output $/1M
$10
$0.87
Blended $/1M (3∶1)
$4
$0.54
Speed
tok/s
95
85
TTFT (s)
0.4
0.4
Reasoning
MMLU-Pro
—
—
GPQA Diamond
91.1
88.8
Humanity's Last Exam
41.3
37.5
AIME 2025
—
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
40.1
43.0
AA-LCR v1.1
82.0
74.7
CritPt
16.9
12.9
MMMU-Pro
77.3
—
IFBench
—
76.5
Chartography
15.6
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
—
—
SWE-bench Multilingual
—
—
LiveCodeBench
—
—
Terminal-Bench 2.1
80.5
64.0
Aider Polyglot
—
—
Terminal-Bench 3
14.6
—
BigCodeBench
—
—
SciCode
54.3
50.8
CursorBench
61.5
—
SWE-Rebench
56.8
41.4
NL2Repo-Bench
—
38.5
DeepSWE
54.0
12.8
WebDev Arena
1,537
1,446
Terminal-Bench 4.0
12.4
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
1,461
1,457
| Metric | Claude Sonnet 5 | DeepSeek V4 Pro | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Anthropic | DeepSeek | — |
| Family | Claude Sonnet / Haiku | DeepSeek | — |
| License | Proprietary | MIT | — |
| Open weights | No | Yes | — |
| Release | Jun 30, 2026 | Apr 24, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | Anthropic | DeepSeek | — |
| Modalities | text, image → text | text → text | — |
| Specs | |||
| Context window | 1M | 1M | tie |
| Max output | 128K | 384K | B +256K |
| Parameters | — | 1.6T (49B act.) | — |
| Pricing | |||
| Input $/1M | $2 | $0.43 | B +$1.56 |
| Output $/1M | $10 | $0.87 | B +$9.13 |
| Blended $/1M (3∶1) | $4 | $0.54 | B +$3.46 |
| Speed | |||
| tok/s | 95 | 85 | A +10 |
| TTFT (s) | 0.4 | 0.4 | tie |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | 91.1 | 88.8 | A +2.3 pts |
| Humanity's Last Exam | 41.3 | 37.5 | A +3.8 pts |
| AIME 2025 | — | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 40.1 | 43.0 | B +2.9 pts |
| AA-LCR v1.1 | 82.0 | 74.7 | A +7.3 pts |
| CritPt | 16.9 | 12.9 | A +4.0 pts |
| MMMU-Pro | 77.3 | — | — |
| IFBench | — | 76.5 | — |
| Chartography | 15.6 | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | — | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | — | — | — |
| Terminal-Bench 2.1 | 80.5 | 64.0 | A +16.5 pts |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | 14.6 | — | — |
| BigCodeBench | — | — | — |
| SciCode | 54.3 | 50.8 | A +3.5 pts |
| CursorBench | 61.5 | — | — |
| SWE-Rebench | 56.8 | 41.4 | A +15.4 pts |
| NL2Repo-Bench | — | 38.5 | — |
| DeepSWE | 54.0 | 12.8 | A +41.2 pts |
| WebDev Arena | 1,537 | 1,446 | A +91 Elo |
| Terminal-Bench 4.0 | 12.4 | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | 1,461 | 1,457 | A +4 Elo |
Claude Sonnet 5
- Context
- 1M / 128K out
- Parameters
- —
- Price
- $2 / $10
- Speed
- 95 tok/s · 0.4s TTFT
- Modalities
- text, image → text
- License
- Proprietary
Default Claude model for Free/Pro at $2/$10 (the launch introductory rate, made permanent on 2026-08-10 - the planned 2026-09-01 increase to $3/$15 was cancelled); closes much of the gap to prior Opus on agentic tasks.
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.