vs
0:2
overall capability wins (1 ties across 3 scored). Claude Sonnet 5.5 leads overall.
- Reasoning0:1
- Coding0:1
Verdict
Claude Sonnet 5.5 leads coding (widest gap: +58.2 on TermBench 4).
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Claude Sonnet 5
- $4.50
- Claude Sonnet 5.5
- $4.50
Price per 1M tokens
Full comparison
Organization
Anthropic
Anthropic
Family
Claude Sonnet / Haiku
Claude Sonnet / Haiku
License
Proprietary
Proprietary
Open weights
No
No
Release
Jun 30, 2026
Sep 28, 2026
Knowledge cutoff
-
-
API / provider
Anthropic
Anthropic
Modalities
text, image → text
text, image → text
Specs
Context window
1M
1M
Max output
128K
-
Parameters
—
—
Pricing
Input $/1M
$2
$2
Output $/1M
$10
$10
Blended $/1M (3∶1)
$4
$4
Speed
tok/s
95
-
TTFT (s)
0.4
-
Reasoning
MMLU-Pro
—
—
GPQA Diamond
91.1
—
Humanity's Last Exam
41.3
—
AIME 2025
—
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
64.5
AA-Omniscience Accuracy
40.1
—
AA-LCR v1.1
82.0
—
CritPt
16.9
—
MMMU-Pro
77.3
—
IFBench
—
—
Chartography
15.6
61.6
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
—
—
SWE-bench Multilingual
—
—
LiveCodeBench
—
—
Terminal-Bench 2.1
80.5
—
Aider Polyglot
—
—
Terminal-Bench 3
14.6
—
BigCodeBench
—
—
SciCode
54.3
—
CursorBench
61.5
—
SWE-Rebench
56.8
—
NL2Repo-Bench
—
—
DeepSWE
54.0
—
WebDev Arena
1,537
—
Terminal-Bench 4.0
12.4
70.6
CursorBench 4.0
—
55.5
FrontierCode v1.1 (Main)
—
52.1
Arena
LMArena Elo
1,461
—
| Metric | Claude Sonnet 5 | Claude Sonnet 5.5 | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Anthropic | Anthropic | — |
| Family | Claude Sonnet / Haiku | Claude Sonnet / Haiku | — |
| License | Proprietary | Proprietary | — |
| Open weights | No | No | — |
| Release | Jun 30, 2026 | Sep 28, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | Anthropic | Anthropic | — |
| Modalities | text, image → text | text, image → text | — |
| Specs | |||
| Context window | 1M | 1M | tie |
| Max output | 128K | - | — |
| Parameters | — | — | — |
| Pricing | |||
| Input $/1M | $2 | $2 | tie |
| Output $/1M | $10 | $10 | tie |
| Blended $/1M (3∶1) | $4 | $4 | tie |
| Speed | |||
| tok/s | 95 | - | — |
| TTFT (s) | 0.4 | - | — |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | 91.1 | — | — |
| Humanity's Last Exam | 41.3 | — | — |
| AIME 2025 | — | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | 64.5 | — |
| AA-Omniscience Accuracy | 40.1 | — | — |
| AA-LCR v1.1 | 82.0 | — | — |
| CritPt | 16.9 | — | — |
| MMMU-Pro | 77.3 | — | — |
| IFBench | — | — | — |
| Chartography | 15.6 | 61.6 | B +46.0 pts |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | — | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | — | — | — |
| Terminal-Bench 2.1 | 80.5 | — | — |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | 14.6 | — | — |
| BigCodeBench | — | — | — |
| SciCode | 54.3 | — | — |
| CursorBench | 61.5 | — | — |
| SWE-Rebench | 56.8 | — | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | 54.0 | — | — |
| WebDev Arena | 1,537 | — | — |
| Terminal-Bench 4.0 | 12.4 | 70.6 | B +58.2 pts |
| CursorBench 4.0 | — | 55.5 | — |
| FrontierCode v1.1 (Main) | — | 52.1 | — |
| Arena | |||
| LMArena Elo | 1,461 | — | — |
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.
Claude Sonnet 5.5
- Context
- 1M
- Parameters
- —
- Price
- $2 / $10
- Speed
- —
- Modalities
- text, image → text
- License
- Proprietary
Anthropic's September 2026 Sonnet model for everyday coding, agentic work, and document creation. It supports text and image input, a 1M context window, and standard API rates of $2/$10 per 1M input/output tokens. Published benchmark results vary by effort and, for agent evaluations, harness; the cited Chartography score is no-tools.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.