vs
0:4
overall capability wins (1 ties across 5 scored). Claude Opus 4.5 leads overall.
- Reasoning0:2
- Coding0:1
- Pricing1:2
- Speed0:2
- Specs0:1
Verdict
Claude Opus 4.5 leads coding (widest gap: +31.1 on SciCode); Claude Opus 4.5 allows ~15.6x more max output (64K vs 4K); Claude Opus 4.5 is ~1.2x cheaper on a blended token basis than Claude 2.1.
- Claude Opus 4.5 leads coding (widest gap: +31.1 on SciCode)
- Claude Opus 4.5 allows ~15.6x more max output (64K vs 4K)
- Claude Opus 4.5 is ~1.2x cheaper on a blended token basis than Claude 2.1
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Claude 2.1
- $14
- Claude Opus 4.5
- $11.25
Claude Opus 4.5 is about 1.2x 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
Nov 21, 2023
Nov 24, 2025
Knowledge cutoff
-
-
API / provider
Anthropic
Anthropic
Modalities
text → text
text, image → text
Specs
Context window
200K
200K
Max output
4K
64K
Parameters
—
—
Pricing
Input $/1M
$8
$5
Output $/1M
$24
$25
Blended $/1M (3∶1)
$12
$10
Speed
tok/s
50
65
TTFT (s)
0.7
0.6
Reasoning
MMLU-Pro
49.5
—
GPQA Diamond
31.9
86.6
Humanity's Last Exam
3.9
30.1
AIME 2025
—
—
MATH-500
37.4
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
—
46.6
AA-LCR v1.1
—
77.3
CritPt
—
4.6
MMMU-Pro
—
74.0
IFBench
—
58.0
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
—
45.9
SWE-bench Multilingual
—
70.7
LiveCodeBench
19.5
—
Terminal-Bench 2.1
—
—
Aider Polyglot
—
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
18.4
49.5
CursorBench
—
—
SWE-Rebench
—
53.1
NL2Repo-Bench
—
—
DeepSWE
—
—
WebDev Arena
—
1,468
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
—
1,470
| Metric | Claude 2.1 | Claude Opus 4.5 | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Anthropic | Anthropic | — |
| Family | Claude Opus | Claude Opus | — |
| License | Proprietary | Proprietary | — |
| Open weights | No | No | — |
| Release | Nov 21, 2023 | Nov 24, 2025 | — |
| Knowledge cutoff | - | - | — |
| API / provider | Anthropic | Anthropic | — |
| Modalities | text → text | text, image → text | — |
| Specs | |||
| Context window | 200K | 200K | tie |
| Max output | 4K | 64K | B +60K |
| Parameters | — | — | — |
| Pricing | |||
| Input $/1M | $8 | $5 | B +$3 |
| Output $/1M | $24 | $25 | A +$1 |
| Blended $/1M (3∶1) | $12 | $10 | B +$2 |
| Speed | |||
| tok/s | 50 | 65 | B +15 |
| TTFT (s) | 0.7 | 0.6 | B +0.10 |
| Reasoning | |||
| MMLU-Pro | 49.5 | — | — |
| GPQA Diamond | 31.9 | 86.6 | B +54.7 pts |
| Humanity's Last Exam | 3.9 | 30.1 | B +26.2 pts |
| AIME 2025 | — | — | — |
| MATH-500 | 37.4 | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | — | 46.6 | — |
| AA-LCR v1.1 | — | 77.3 | — |
| CritPt | — | 4.6 | — |
| MMMU-Pro | — | 74.0 | — |
| IFBench | — | 58.0 | — |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | — | 45.9 | — |
| SWE-bench Multilingual | — | 70.7 | — |
| LiveCodeBench | 19.5 | — | — |
| Terminal-Bench 2.1 | — | — | — |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | 18.4 | 49.5 | B +31.1 pts |
| CursorBench | — | — | — |
| SWE-Rebench | — | 53.1 | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | — | — |
| WebDev Arena | — | 1,468 | — |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | — | 1,470 | — |
Claude 2.1
- Context
- 200K / 4K out
- Parameters
- —
- Price
- $8 / $24
- Speed
- 50 tok/s · 0.7s TTFT
- Modalities
- text → text
- License
- Proprietary
Pre-Claude-3 long-context model (200K) — historical baseline for early RAG and document Q&A.
Claude Opus 4.5
- Context
- 200K / 64K out
- Parameters
- —
- Price
- $5 / $25
- Speed
- 65 tok/s · 0.6s TTFT
- Modalities
- text, image → text
- License
- Proprietary
Late-2025 Opus release that set a high bar on SWE-bench; superseded by 4.8 and Opus 5 for new projects.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.