Claude Opus 5.5
Anthropic · released
The model Anthropic tells most developers to start with: built for long autonomous coding sessions and document-heavy professional work. It lands close to — and on many benchmarks above — the pricier Fable 5.1 at 40% of the per-token price.
Best for: Agentic coding · Knowledge work · Computer use
- Intelligence Index
- 57.6
- #1 of 38 · at Max effort
- Price per 1M tokens
- $4 in $20 out
- Blended $8 at 3:1
- Context window
- 1M
- Up to 128K output
- Output speed
- 95 tokens/s
- $5.98 measured cost per task
Capability profile
One 0–100 score per area, placing this model between the weakest and strongest tracked model across that area's benchmarks.
- Overall96 · #1 of 38
Based on 3 of 3 benchmarks
- Coding97 · #1 of 39
Based on 5 of 5 benchmarks
- Agents & knowledge work91 · #2 of 40
Based on 6 of 6 benchmarks
- Reasoning & science100 · #1 of 39
Based on 2 of 2 benchmarks
- Knowledge & reliability97 · #1 of 39
Based on 2 of 2 benchmarks
- Vision100 · #1 of 31
Based on 1 of 2 benchmarks
Reasoning effort trade-off
The same model at each effort setting. Lower settings give up a few points for a fraction of the cost and wait.
| Setting | Intelligence Index | Cost / task | First answer |
|---|---|---|---|
| Max | 57.6 | $5.98 | 12 min |
| xHigh | 56.0 | $3.46 | 2.2 min |
| HighSweet spot | 53.6 | $1.82 | 38s |
| Medium | 51.2 | $1.34 | 23s |
| Low | 42.3 | $0.55 | 8.7s |
Measured by Artificial Analysis. First answer is the median wait until the first answer token, including thinking time. “Sweet spot” is the cheapest setting that keeps at least 90% of the top score for no more than 60% of its cost.
Benchmark results
All third-party results on record. Rank is among the tracked models that have been tested on that benchmark.
Overall
#1 of 38· Max effort· Artificial Analysis
- Vals Index67.0%
#3 of 26· Max effort· $32.1 per test· Vals AI
- Arena · Text1504 ±9
#2 of 33· High effort· 4,552 votes· Arena
Coding
- Terminal-Bench 4.059.6%
#2 of 38· Max effort· Artificial Analysis
- SciCode66.9%
#1 of 38· Max effort· Artificial Analysis
- Arena · WebDev1814 ±16
#1 of 34· Max effort· 2,229 votes· Arena
- Vibe Code Bench90.3%
#4 of 26· Max effort· Vals AI
- Code Migration68.9%
#3 of 26· Max effort· Vals AI
Agents & knowledge work
- GDPval-AA68.3
#1 of 39· Max effort· Artificial Analysis
- Arena · Agent+13.8 ±2.2
#2 of 29· High effort· Arena
- Finance Agent v258.6%
#5 of 26· Max effort· Vals AI
#2 of 26· Max effort· Vals AI
- Legal Research Bench50.5%
#4 of 26· Max effort· Vals AI
- Tax Agent Bench70.5%
#6 of 26· Max effort· Vals AI
Reasoning & science
- Humanity's Last Exam61.4%
#1 of 39· Max effort· Artificial Analysis
- CritPt31.7%
#1 of 39· Max effort· Artificial Analysis
- ARC-AGI-2saturating93.3%
#3 of 16· High effort· $0.41 per task· ARC Prize
- ARC-AGI-3Not tested
- GPQA DiamondsaturatingNot tested
Knowledge & reliability
#1 of 39· Max effort· Artificial Analysis
- AA-LCR84.7%
#5 of 39· Max effort· Artificial Analysis
Vision
- MMMU-Pro87.7%
#1 of 27· Max effort· Artificial Analysis
- Arena · VisionNot tested
Reported by Anthropic
Numbers the lab published at launch, run on its own harness and settings. Useful context, but not independently verified and never used in the rankings on this site.
- Terminal-Bench 4.0 · xHigh effort66.4%
- SWE-bench Pro · Max effort89.9%
- OSWorld 2.1 · partial score81.8%
- CursorBench 4.0 · run by Cursor57.8%
- Humanity's Last Exam · no tools64.4%
Specifications
- Lab
- Anthropic
- Released
- Sep 22, 2026
- Availability
- Generally available
- Context window
- 1,000,000 tokens
- Max output
- 128,000 tokens
- Accepts
- Text, Images, PDFs
- Knowledge cutoff
- Jun 2026
- Weights
- Closed
- Reasoning control
- Always on · effort low to max
- API model ID
claude-opus-5-5
Pricing
- Input / 1M tokens
- $4
- Cached input / 1M
- $0.20
- Output / 1M tokens
- $20
- Blended (3:1)
- $8
- Measured cost / task
- $5.98
Full 1M context at standard rates. Batch API is half price; a fast mode costs double.
Worth knowing
- Safeguards hand some cybersecurity and life-science requests to an older Opus model. Independent evaluators score what the product returns, so their results include those fallbacks.
Official sources
Closest alternatives
The models nearest to this one on overall score.
23 benchmarks tracked. See methodology and sources.