vs
2:8
overall capability wins (2 ties across 12 scored). Claude Opus 4.8 leads overall.
- Reasoning1:5
- Coding0:3
- Arena1:0
- Speed0:2
Verdict
Claude Opus 4.8 leads coding (widest gap: +3.2 on SWE-Rebench); Claude Opus 4.6 ranks higher on LMArena (+24 Elo).
- Claude Opus 4.8 leads coding (widest gap: +3.2 on SWE-Rebench)
- Claude Opus 4.6 ranks higher on LMArena (+24 Elo)
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Claude Opus 4.6
- $11.25
- Claude Opus 4.8
- $11.25
Price per 1M tokens
Full comparison
Organization
Anthropic
Anthropic
Family
Claude Opus
Claude Opus
License
Proprietary
Proprietary
Open weights
No
No
Release
Feb 19, 2026
May 28, 2026
Knowledge cutoff
-
-
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
$5
Output $/1M
$25
$25
Blended $/1M (3∶1)
$10
$10
Speed
tok/s
65
72
TTFT (s)
0.6
0.55
Reasoning
MMLU-Pro
—
—
GPQA Diamond
89.6
92.0
Humanity's Last Exam
39.9
48.7
AIME 2025
—
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
47.0
48.8
AA-LCR v1.1
78.0
77.7
CritPt
12.6
20.9
MMMU-Pro
75.4
—
IFBench
53.1
62.2
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
75.6
—
SWE-bench Pro
—
69.2
SWE-bench Multilingual
72.0
—
LiveCodeBench
—
—
Terminal-Bench 2.1
—
84.6
Aider Polyglot
—
—
Terminal-Bench 3
—
21.1
BigCodeBench
—
—
SciCode
51.9
54.4
CursorBench
—
62.3
SWE-Rebench
53.3
56.5
NL2Repo-Bench
—
—
DeepSWE
—
59.0
WebDev Arena
1,537
1,539
Terminal-Bench 4.0
—
23.6
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
1,497
1,473
| Metric | Claude Opus 4.6 | Claude Opus 4.8 | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Anthropic | Anthropic | — |
| Family | Claude Opus | Claude Opus | — |
| License | Proprietary | Proprietary | — |
| Open weights | No | No | — |
| Release | Feb 19, 2026 | May 28, 2026 | — |
| Knowledge cutoff | - | - | — |
| 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 | $5 | tie |
| Output $/1M | $25 | $25 | tie |
| Blended $/1M (3∶1) | $10 | $10 | tie |
| Speed | |||
| tok/s | 65 | 72 | B +7.00 |
| TTFT (s) | 0.6 | 0.55 | B +0.05 |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | 89.6 | 92.0 | B +2.4 pts |
| Humanity's Last Exam | 39.9 | 48.7 | B +8.8 pts |
| AIME 2025 | — | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 47.0 | 48.8 | B +1.8 pts |
| AA-LCR v1.1 | 78.0 | 77.7 | A +0.3 pts |
| CritPt | 12.6 | 20.9 | B +8.3 pts |
| MMMU-Pro | 75.4 | — | — |
| IFBench | 53.1 | 62.2 | B +9.1 pts |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | 75.6 | — | — |
| SWE-bench Pro | — | 69.2 | — |
| SWE-bench Multilingual | 72.0 | — | — |
| LiveCodeBench | — | — | — |
| Terminal-Bench 2.1 | — | 84.6 | — |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | 21.1 | — |
| BigCodeBench | — | — | — |
| SciCode | 51.9 | 54.4 | B +2.5 pts |
| CursorBench | — | 62.3 | — |
| SWE-Rebench | 53.3 | 56.5 | B +3.2 pts |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | 59.0 | — |
| WebDev Arena | 1,537 | 1,539 | B +2 Elo |
| Terminal-Bench 4.0 | — | 23.6 | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | 1,497 | 1,473 | A +24 Elo |
Claude Opus 4.6
- Context
- 1M / 128K out
- Parameters
- —
- Price
- $5 / $25
- Speed
- 65 tok/s · 0.6s TTFT
- Modalities
- text, image → text
- License
- Proprietary
Mid-stack Opus generation between 4.5 and 4.7 - strong coding and long context.
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.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.