Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Anthropic announced Claude Sonnet 4.5 on September 29, 2025, positioning it as a model for coding, complex agents, computer use, reasoning, and math. Its launch benchmarks offer useful evidence about specific test setups, not a promise of performance on your own work. And because Anthropic’s current Sonnet family page now lists newer generations, check whether Sonnet 4.5 is still offered in the app or API channel you plan to use.
What is Claude Sonnet 4.5?
Claude Sonnet 4.5 is a model in Anthropic’s Claude family, announced on September 29, 2025. At launch, Anthropic highlighted coding, sustained multi-step work, computer use, reasoning, and math as strengths. The model was presented as suitable for both interactive use and agent-style workflows that can take multiple actions.
As an Amazon Associate I earn from qualifying purchases.
Anthropic also announced product and developer-tool updates alongside the model: checkpoints and a refreshed terminal interface for Claude Code, a native VS Code extension, context editing and a memory tool for the Claude API, code execution and file creation in Claude apps, Claude for Chrome access for some Max users, and the Claude Agent SDK. These are launch-era announcements, not a guarantee that each feature or entitlement remains available on every current plan.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minuteAnthropic described a browser demonstration in which the model navigated a site and completed a spreadsheet. That illustrates the kind of computer-use workflow the company targeted, but a demonstration does not establish reliability across arbitrary websites or desktop tasks.
Is Claude Sonnet 4.5 good for coding?
Anthropic reported a 77.2% score on SWE-bench Verified, a benchmark of real software issues. The company averaged 10 trials on the 500-problem set, using a simple bash and file-editing scaffold, no test-time compute, and a 200K thinking budget. Those conditions matter: the result describes performance under that evaluation setup, not a predicted success rate for your repository or coding workflow.
Anthropic also reported an 82.0% SWE-bench Verified result in a separate high-compute setup. That evaluation used multiple parallel attempts, rejected patches that broke visible regression tests, and an internal scoring model to select a candidate. It is not directly comparable to the 77.2% result because the methods differ.
Anthropic’s launch announcement contrasted Sonnet 4.5’s 61.4% on OSWorld-Verified with 42.2% for Sonnet 4 on OSWorld. Sonnet 4.5’s result was averaged across four runs using the official framework with a 100-step maximum. Because the announcement identifies different benchmark versions, treat the comparison as suggestive rather than a perfectly controlled, apples-to-apples improvement.
The company also said Sonnet 4.5 could focus for more than 30 hours on complex, multi-step tasks. That is Anthropic’s observation, not a standardized public benchmark figure. Customer comments in the announcement—including statements from Cursor CEO Michael Truell and GitHub Chief Product Officer Mario Rodriguez—are attributed customer assessments, not independent validation.
For an actual coding decision, test the model on representative issues from your own codebase. Check whether it understands project conventions, produces patches that pass your test suite, handles follow-up changes, and recovers cleanly when a tool call fails. Benchmark scores can help narrow choices, but your task mix and evaluation method determine whether the model is useful to your team.
What did Anthropic say about computer use and safety?
Anthropic said Sonnet 4.5 launched under its AI Safety Level 3 protections, including classifiers intended to detect potentially dangerous inputs and outputs. The company also acknowledged that classifiers can incorrectly flag benign content. These safeguards are not a guarantee that every action will be safe or correct.
Computer-using agents face prompt-injection risks: untrusted content on a web page or in a document may try to redirect an agent or induce unwanted actions. Anthropic said it had improved prompt-injection defenses while recognizing the risk as serious. For workflows that can send messages, change records, access sensitive files, or make purchases, constrain permissions and require human review for consequential actions. Keep a way to inspect, stop, and recover from an agent’s work.
How much does Claude Sonnet 4.5 cost?
Anthropic’s September 2025 announcement listed API launch prices of $3 per million input tokens and $15 per million output tokens. Those are historical launch prices, not a current quote. Current pricing, usage limits, and availability were not established by the launch announcement; check Anthropic’s current API pricing and model documentation before budgeting.
Token rates alone do not determine the cost of an agent workflow. Repeated attempts, long context, tool use, and the amount of output can change total spend. When comparing models, account for current input and output rates, context limits, caching or batch options, regional charges where applicable, and the cost of runs that fail or need human correction.
Is Claude Sonnet 4.5 still available?
Anthropic’s current Sonnet family page lists Sonnet 4.6, Sonnet 5, and Sonnet 5.5 after Sonnet 4.5. That catalog does not by itself establish whether Sonnet 4.5 remains selectable in every Claude app, API, or cloud-provider channel. Verify the specific channel and account you intend to use before building around it; model names and access routes can change.
If Sonnet 4.5 is available to you, compare it with currently offered alternatives on the tasks you actually need. Check the model generation and channel, current price and limits, task performance on your own examples, latency and completion cost, agent permissions, and failure recovery. Do not select a model solely because of a launch-era benchmark or price.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
