vs
0:4
overall capability wins (2 ties across 6 scored). Claude Opus 5.5 leads overall.
- Reasoning0:1
- Coding0:2
- Agents0:1
- Pricing0:3
Verdict
Claude Opus 5.5 leads coding (widest gap: +42.8 on TermBench 4); Claude Opus 5.5 is ~1.3x cheaper on a blended token basis than Claude Opus 4.8.
- Claude Opus 5.5 leads coding (widest gap: +42.8 on TermBench 4)
- Claude Opus 5.5 is ~1.3x cheaper on a blended token basis than Claude Opus 4.8
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Claude Opus 4.8
- $11.25
- Claude Opus 5.5
- $9
Claude Opus 5.5 is about 1.3x cheaper on this workload.
Price per 1M tokens
Full comparison
Organization
Anthropic
Anthropic
Family
Claude Opus
Claude Opus
License
Proprietary
Proprietary
Open weights
No
No
Release
May 28, 2026
Sep 22, 2026
Knowledge cutoff
-
2026-06
API / provider
Anthropic
Anthropic
Modalities
text, image → text
text, image → text
Specs
Context window
1M
1M
Max output
128K
128K
Parameters
—
—
Pricing
Input $/1M
$5
$4
Output $/1M
$25
$20
Blended $/1M (3∶1)
$10
$8
Speed
tok/s
72
-
TTFT (s)
0.55
-
Reasoning
MMLU-Pro
—
—
GPQA Diamond
92.0
—
Humanity's Last Exam
48.7
61.4
AIME 2025
—
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
67.7
AA-Omniscience Accuracy
48.8
—
AA-LCR v1.1
77.7
—
CritPt
20.9
—
MMMU-Pro
—
—
IFBench
62.2
—
Chartography
—
64.4
Chartography (With Tools)
—
89.0
Coding
SWE-bench Verified
—
—
SWE-bench Pro
69.2
—
SWE-bench Multilingual
—
—
LiveCodeBench
—
—
Terminal-Bench 2.1
84.6
—
Aider Polyglot
—
—
Terminal-Bench 3
21.1
—
BigCodeBench
—
—
SciCode
54.4
66.9
CursorBench
62.3
—
SWE-Rebench
56.5
—
NL2Repo-Bench
—
—
DeepSWE
59.0
—
WebDev Arena
1,539
—
Terminal-Bench 4.0
23.6
66.4
CursorBench 4.0
—
57.8
FrontierCode v1.1 (Main)
—
54.4
Arena
LMArena Elo
1,473
—
| Metric | Claude Opus 4.8 | Claude Opus 5.5 | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Anthropic | Anthropic | — |
| Family | Claude Opus | Claude Opus | — |
| License | Proprietary | Proprietary | — |
| Open weights | No | No | — |
| Release | May 28, 2026 | Sep 22, 2026 | — |
| Knowledge cutoff | - | 2026-06 | — |
| API / provider | Anthropic | Anthropic | — |
| Modalities | text, image → text | text, image → text | — |
| Specs | |||
| Context window | 1M | 1M | tie |
| Max output | 128K | 128K | tie |
| Parameters | — | — | — |
| Pricing | |||
| Input $/1M | $5 | $4 | B +$1 |
| Output $/1M | $25 | $20 | B +$5 |
| Blended $/1M (3∶1) | $10 | $8 | B +$2 |
| Speed | |||
| tok/s | 72 | - | — |
| TTFT (s) | 0.55 | - | — |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | 92.0 | — | — |
| Humanity's Last Exam | 48.7 | 61.4 | B +12.7 pts |
| AIME 2025 | — | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | 67.7 | — |
| AA-Omniscience Accuracy | 48.8 | — | — |
| AA-LCR v1.1 | 77.7 | — | — |
| CritPt | 20.9 | — | — |
| MMMU-Pro | — | — | — |
| IFBench | 62.2 | — | — |
| Chartography | — | 64.4 | — |
| Chartography (With Tools) | — | 89.0 | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | 69.2 | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | — | — | — |
| Terminal-Bench 2.1 | 84.6 | — | — |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | 21.1 | — | — |
| BigCodeBench | — | — | — |
| SciCode | 54.4 | 66.9 | B +12.5 pts |
| CursorBench | 62.3 | — | — |
| SWE-Rebench | 56.5 | — | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | 59.0 | — | — |
| WebDev Arena | 1,539 | — | — |
| Terminal-Bench 4.0 | 23.6 | 66.4 | B +42.8 pts |
| CursorBench 4.0 | — | 57.8 | — |
| FrontierCode v1.1 (Main) | — | 54.4 | — |
| Arena | |||
| LMArena Elo | 1,473 | — | — |
Claude Opus 4.8
- Context
- 1M / 128K out
- Parameters
- —
- Price
- $5 / $25
- Speed
- 72 tok/s · 0.55s TTFT
- Modalities
- text, image → text
- License
- Proprietary
Previous Opus flagship with 1M context; still competitive for coding agents and computer use.
Claude Opus 5.5
- Context
- 1M / 128K out
- Parameters
- —
- Price
- $4 / $20
- Speed
- —
- Modalities
- text, image → text
- License
- Proprietary
Anthropic's September 2026 Opus model for long-running agentic coding and knowledge work. It has a 1M context window, 128K output limit, always-on adaptive thinking, and standard API rates of $4/$20 per 1M input/output tokens.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.