benchmarks
Claude Opus 4.5 is an Anthropic model designed for software engineering, agentic workflows, and computer use, with published standard API pricing of $5 per million input tokens and $25 per million output tokens. Current availability, retirement status, and subsequent releases: Not yet confirmed by public reporting.
Benchmark results
Every figure links to where it was published. Compare like with like — harnesses and prompt setups differ between vendors.
| Benchmark | Score | Source |
|---|---|---|
| GPQA DiamondPhD-level science questions | 87%24 Nov 2025 | View |
| SWE-bench VerifiedResolving real GitHub issues | 80.9%24 Nov 2025 | View |
| Terminal-BenchLong-horizon command-line agent tasks | 59.3%24 Nov 2025 | View |
| OTIS Mock AIME 2024-25Competition mathematics | 86.1%24 Nov 2025 | View |
| FrontierMath | 20.7%25 Nov 2025 | View |
| SimpleQA VerifiedShort factual questions (hallucination resistance) | 45.7%27 Aug 2026 | View |
| LMArena TextHuman preference Elo from blind chat votes (not a percentage) | 1451 Elo22 Oct 2026 | View |
| LMArena WebDevHuman preference Elo for building web apps (not a percentage) | 1470 Elo23 Oct 2026 | View |
Weighted mean of 6 verified benchmark results: GPQA Diamond 87%, SWE-bench Verified 80.9%, Terminal-Bench 59.3%, OTIS Mock AIME 2024-25 86.1%, FrontierMath 20.7%, SimpleQA Verified 45.7%.
Capabilities
- Repository-level software engineering, including debugging and code changes.
- Multi-step agentic workflows involving tools and external applications.
- Computer-use tasks through supported agent integrations.
- Analysis and synthesis of documents within a 200K-token context window.
- Configurable effort to balance response quality, latency, and token consumption.
What improved
- Anthropic reported stronger software-engineering performance than Claude Opus 4.1.
- Maximum output increased from Opus 4.1's 32K tokens to 64K tokens.
- Standard input pricing decreased from $15 to $5 per million tokens compared with Opus 4.1.
- Standard output pricing decreased from $75 to $25 per million tokens compared with Opus 4.1.
Known flaws & limitations
- The 200K-token context window limits the material available in an individual request.
- The 64K-token maximum output limits the length of an individual response.
- Published benchmark results depend on evaluation settings and do not guarantee equivalent production performance.
- Generated code and tool actions require validation; strong benchmark performance does not establish correctness or security.
Risks, safety & security issues
- Untrusted webpages, documents, and tool outputs can expose agents to prompt-injection attempts.
- Broad tool permissions can amplify the consequences of incorrect or manipulated actions.
- Software-engineering capabilities are dual-use and can support harmful development.
- Sensitive information supplied to tool-enabled workflows can be exposed through inappropriate downstream actions.
Good fit for
- Supervised coding, debugging, and repository maintenance.
- Multi-step research and document synthesis within the context limit.
- Tool-assisted business workflows with scoped permissions and approval gates.
- Complex technical tasks where output can be independently tested.
Poor fit for
- Unsupervised actions involving production credentials, irreversible changes, or financial transfers.
- High-stakes decisions without qualified human review.
- Workloads requiring more than 200K tokens in one request or more than 64K output tokens.
- Security-critical code deployment without testing and review.
References
Profile last reviewed 10/27/2026 and re-checked automatically every few hours against the vendor's published documentation, papers and public leaderboards. Scores come from the cited benchmark results above — confirm limits and pricing against the provider's own documentation before building on them.