Anthropic launches Claude Haiku 4.5: a small AI model that delivers Sonnet-4-level encoding performance at one-third the cost and more than twice the speed

Anthropic release Claude Haiku 4.5a latency-optimized “small” model that can provide Encoding performance levels similar to Claude Sonnet 4 while running More than twice the speed, one-third less cost. The model is immediately available through Anthropic’s API and partner directory amazon bedrock and Google Cloud Apex Artificial Intelligence. Pricing is $1/MTok input and $5/MTok output. Anthropic positions Haiku 4.5 as replacement Haiku 3.5 and Sonnet 4 In cost-sensitive interactive workloads.

Positioning and lineup

Haiku 4.5 targets real-time assistants, customer support automation, and pair programming, where latency budgeting and throughput dominate. it Outperforms Sonnet 4 in Computer Usage task– GUI/browser operation underpins products such as Claude for Chrome – and is described as significantly improving responsiveness Claude Corder Used for multi-agent projects and rapid prototyping. anthropic clearly shows Sonnet 4.5 remains a cutting-edge model and “the best coding model in the world” And Haiku 4.5 provides Near-leading performance and greater cost-effectiveness. The recommended pattern is Sonnet 4.5 for multi-step planning and Executed in parallel by Haiku 4.5 worker thread pool.

Availability, identifiers and pricing

From day one, developers can call the model (claude-haiku-4-5) on Anthropic’s API. Anthropic also states availability amazon bedrock and Apex Artificial Intelligence;The model catalog may update regional coverage and IDs over time, but the company confirmed cloud availability in a release post. this API price For Haiku 4.5 it is $1/MTok (input) and $5/MTok (output)prompting the cache to be listed in $1.25/MTok write and $0.10/MTok Read.

Benchmark

Anthropic summarizes the results for the standard suite and the surrogate suite, and includes details of the methodology used to verify the numbers:

• SWE-bench verified: Simple scaffolding with two tools (bash, file editing), 73.3% average over 50 trials, Calculated when there are no tests, 128K Thinking Budgetdefault sampling. Includes a small appendix of tips to encourage extensive use of the tool and writing tests first.
• terminal bench: Terminus-2 agent, averaged over 11 runs (6 without thinking, 5 with thinking) 32K Thinking Budget).
• OSWorld Verification: Up to 100 stepsthe average of 4 runs 128K total thinking budget and 2K per step configuration.
• Aimee/MMMLU:Average multiple runs using default sampling 128K Thinking Budget.

Post highlights Coding parity with Sonnet 4 and Computer usage revenue Relative to these brackets are Sonnet 4. Users should replicate using their own orchestration, tool stack, and mental budget before generalizing.

Main points

• Haiku 4.5 delivers Sonnet-4-level encoding performance at one-third the cost and more than twice the speed.
• It surpasses Sonnet 4 in computer usage tasks, improving the responsiveness of Claude for Chrome and the multi-agent flow of Claude Code.
• Recommended orchestration: Use Sonnet 4.5 for multi-step planning and use multiple Haiku 4.5 worker threads for parallel execution.
• Pricing is $1/5 per million input/output tokens; available via Claude API, Amazon Bedrock, and Google Cloud Vertex AI.
• Released under ASL-2, misalignment rates measured in Anthropic’s tests were lower than those of Sonnet 4.5 and Opus 4.1.

Anthropic’s positioning of Claude Haiku 4.5 makes strategic sense: by delivering Encoding performance levels similar to Claude Sonnet 4 exist one third of the cost and More than twice the speedalthough Surpasses Sonnet 4 in computer usagethe company provides developers with a clear division between planners and executors——Sonnet 4.5 for multi-step planning and pooling Haiku 4.5 Parallel execution of workers – no forced schema changes (“direct replacement“Through API, Amazon Bedrock, Vertex AI). this ASL-2 Published with records Lower misalignment rate Compare Sonnet 4.5 and Work 4.1reduces the friction of enterprise deployments, where security gates and cost envelopes dominate the deployment math.

Check Technical details, system cards, model pages and documentation . Please feel free to check out our GitHub page for tutorials, code, and notebooks. In addition, welcome to follow us twitter And don’t forget to join our 100k+ ML SubReddit and subscribe our newsletter. wait! Are you using Telegram? Now you can also join us via telegram.

Asif Razzaq is the CEO of Marktechpost Media Inc. As a visionary entrepreneur and engineer, Asif is committed to harnessing the potential of artificial intelligence for the benefit of society. His most recent endeavor is the launch of Marktechpost, an artificial intelligence media platform that stands out for its in-depth coverage of machine learning and deep learning news that is technically sound and easy to understand for a broad audience. The platform has more than 2 million monthly views, which shows that it is very popular among viewers.