Claude Sonnet 4.5
Anthropic's balanced Claude model with strong reasoning and efficiency.
About Claude Sonnet 4.5
Claude Sonnet 4.5 was built with software engineers in mind — it set a record of 77.2% on SWE-bench Verified at launch, the highest any model had achieved on that benchmark, and can sustain autonomous coding sessions across dozens of turns without losing context. Extended thinking support lets it trade response speed for deeper multi-step reasoning when accuracy matters. Developers praise its ability to navigate real codebases, reproduce bugs reliably, and orchestrate tools across long-horizon tasks. The trade-off is real: output speed sits at 40.9 tokens per second, meaningfully below the peer median of 61.3, and a vocal portion of developers report that benchmark-topping scores don't always carry over to everyday coding workflows — some note persistent instruction-following lapses that its predecessors handled more consistently. Now in legacy status as of mid-2026, Sonnet 4.5 remains fully supported and is worth considering for teams with established agentic pipelines or budget-conscious production workloads that don't yet need a Sonnet 4.6 upgrade.
Best for
- Autonomous software engineering and bug-fixing on real GitHub codebases
- Long-horizon agentic coding workflows requiring sustained context over many turns
- Computer use and desktop automation tasks (61.4% on OSWorld)
- Complex reasoning tasks where extended thinking mode can be enabled for greater accuracy
- Cost-effective production inference compared to Opus-tier models
Specs & capabilities
How Claude Sonnet 4.5 stacks up — intelligence, speed, context, and modalities.
Intelligence
High
Speed
Slow
Context window
200,000 tokens
Max output
64,000 tokens
Knowledge cutoff
January 2025
Frequently asked questions
What does Claude Sonnet 4.5 cost?
Input is $3.00 per million tokens and output is $15.00 per million tokens. Batch API discounts cut those rates in half ($1.50/$7.50). Prompt caching is available at $0.30 per million tokens for cache hits.
What is the context window?
200,000 tokens by default, with up to 1,000,000 tokens available via a beta flag. Maximum output per response is 64,000 tokens — shorter than some predecessors.
Is this model still actively maintained?
Sonnet 4.5 is listed in Anthropic's legacy models section as of mid-2026. It remains fully supported and available for production use but has been superseded by Claude Sonnet 4.6.
What is it best at?
Software engineering. It achieved 77.2% on SWE-bench Verified at launch — a benchmark that measures solving real GitHub issues — and includes strong computer use performance at 61.4% on OSWorld.
What are the known weaknesses?
Output generation is notably slow at around 40.9 tokens per second. Some developers also report a gap between benchmark scores and real-world workflow reliability, including occasional instruction-following lapses.
How does it compare to Claude Sonnet 4.6?
Sonnet 4.6 is Anthropic's current mid-tier model and offers improved performance and reasoning over 4.5. Sonnet 4.5 may still be preferable for teams with existing pipelines tuned to its behavior or those not yet ready to migrate.