vs
0:4
overall capability wins (1 ties across 5 scored). Claude Sonnet 4.6 leads overall.
- Coding0:2
- Arena0:1
- Speed0:1
- Specs0:1
Verdict
Claude Sonnet 4.6 leads coding (widest gap: +135.5 on WebDev Elo); Claude Sonnet 4.6 ranks higher on LMArena (+18 Elo); Claude Sonnet 4.6 offers ~5.0x larger context (1M vs 200K).
- Claude Sonnet 4.6 leads coding (widest gap: +135.5 on WebDev Elo)
- Claude Sonnet 4.6 ranks higher on LMArena (+18 Elo)
- Claude Sonnet 4.6 offers ~5.0x larger context (1M vs 200K)
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Claude Sonnet 4.5
- $6.75
- Claude Sonnet 4.6
- $6.75
Price per 1M tokens
Full comparison
Organization
Anthropic
Anthropic
Family
Claude Sonnet / Haiku
Claude Sonnet / Haiku
License
Proprietary
Proprietary
Open weights
No
No
Release
Sep 29, 2025
Mar 18, 2026
Knowledge cutoff
-
-
API / provider
Anthropic
Anthropic
Modalities
text, image → text
text, image → text
Specs
Context window
200K
1M
Max output
64K
64K
Parameters
—
—
Pricing
Input $/1M
$3
$3
Output $/1M
$15
$15
Blended $/1M (3∶1)
$6
$6
Speed
tok/s
85
90
TTFT (s)
0.4
0.4
Reasoning
MMLU-Pro
—
—
GPQA Diamond
—
87.5
Humanity's Last Exam
—
33.6
AIME 2025
—
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
—
40.9
AA-LCR v1.1
—
80.0
CritPt
—
3.1
MMMU-Pro
—
73.3
IFBench
—
56.6
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
43.6
—
SWE-bench Multilingual
67.0
—
LiveCodeBench
—
—
Terminal-Bench 2.1
—
71.2
Aider Polyglot
—
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
—
50.1
CursorBench
—
—
SWE-Rebench
52.4
54.5
NL2Repo-Bench
—
—
DeepSWE
—
—
WebDev Arena
1,386
1,521
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
1,455
1,473
| Metric | Claude Sonnet 4.5 | Claude Sonnet 4.6 | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Anthropic | Anthropic | — |
| Family | Claude Sonnet / Haiku | Claude Sonnet / Haiku | — |
| License | Proprietary | Proprietary | — |
| Open weights | No | No | — |
| Release | Sep 29, 2025 | Mar 18, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | Anthropic | Anthropic | — |
| Modalities | text, image → text | text, image → text | — |
| Specs | |||
| Context window | 200K | 1M | B +800K |
| Max output | 64K | 64K | tie |
| Parameters | — | — | — |
| Pricing | |||
| Input $/1M | $3 | $3 | tie |
| Output $/1M | $15 | $15 | tie |
| Blended $/1M (3∶1) | $6 | $6 | tie |
| Speed | |||
| tok/s | 85 | 90 | B +5.00 |
| TTFT (s) | 0.4 | 0.4 | tie |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | — | 87.5 | — |
| Humanity's Last Exam | — | 33.6 | — |
| AIME 2025 | — | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | — | 40.9 | — |
| AA-LCR v1.1 | — | 80.0 | — |
| CritPt | — | 3.1 | — |
| MMMU-Pro | — | 73.3 | — |
| IFBench | — | 56.6 | — |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | 43.6 | — | — |
| SWE-bench Multilingual | 67.0 | — | — |
| LiveCodeBench | — | — | — |
| Terminal-Bench 2.1 | — | 71.2 | — |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | — | 50.1 | — |
| CursorBench | — | — | — |
| SWE-Rebench | 52.4 | 54.5 | B +2.1 pts |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | — | — |
| WebDev Arena | 1,386 | 1,521 | B +136 Elo |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | 1,455 | 1,473 | B +18 Elo |
Claude Sonnet 4.5
- Context
- 200K / 64K out
- Parameters
- —
- Price
- $3 / $15
- Speed
- 85 tok/s · 0.4s TTFT
- Modalities
- text, image → text
- License
- Proprietary
Prior Sonnet workhorse for coding agents before the 4.6 / 5 line.
Claude Sonnet 4.6
- Context
- 1M / 64K out
- Parameters
- —
- Price
- $3 / $15
- Speed
- 90 tok/s · 0.4s TTFT
- Modalities
- text, image → text
- License
- Proprietary
Prior Sonnet generation with strong coding and long-context support before Sonnet 5.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.