Anthropic Launches Claude Haiku 4.5: Small AI Model that Delivers Sonnet-4-Level Coding Performance at One-Third the Cost and more than Twice the Speed

Anthropic released Claude Haiku 4.5a latency-optimized “small” model that delivers good performance Similar levels of programming performance to Cloud Sonet 4 While running More than twice the speed at a third of the cost. The model is immediately available via Anthropic’s API and in partner catalogs at Amazon Bedrock and Google Cloud Vertex AI. Pricing is Input $1/MTok and output $5/MTok. Human Attitudes Haiku 4.5 as a direct replacement for Haiku 3.5 and Sonnet 4 In interactive and cost-sensitive workloads.

Positioning and formation

Haiku 4.5 targets real-time assistants, customer support automation, and binary programming where tight latency and throughput budgets dominate. He – she It outperforms the Sonnet 4 at “computer usage” tasks.– GUI/browser manipulation that supports products like Claude for Chrome – and is described as significantly improving responsiveness in Claude Code For multi-agent projects and rapid prototyping. Anthropy explains this Sonnet 4.5 remains the leading model and “the best coding model in the world”. While Haiku offers 4.5 Close to the limit performance with greater cost effectiveness. The recommended style is Sonnet 4.5 for multi-step planning and Parallel execution by a group of Haiku workers 4.5.

Availability, identifiers and pricing

From day one, developers can call the form (claude-haiku-4-5) on the Anthropic API. Anthropy also states availability on Amazon Bedrock and Vertex Artificial Intelligence; Model catalogs may update region coverage and identifiers over time, but the company confirms cloud availability in the launch post. the API price Haiku 4.5 is $1/MTok (input) and $5/MTok(output)with the inclusion of fast caching in $1.25/MTok write and Read $0.10/MTok.

Standards

Anthropic summarizes the results across the standard and proxy wings and includes details of the methodology for qualifying the numbers:

• SWE seat verified: A simple scaffold with two tools (bash, edit files), 73.3% On average more than 50 trials, There is no test time calculation, 128 thousand thinking budget,Virtual sampling. Includes a simple quick appendix that encourages widespread use of the tool and writing tests first.
• Bench station: Terminus Agent-2, average over 11 rounds (6 without thinking, 5 with 32 thousand budget thinking).
• Verified by OSWorld: 100 steps maximumaveraging 4 runs with A Total thinking budget is 128k and 2K per step Settings.
• Amy/Mamlo: Averages over multiple runs using default sampling and 128 thousand thinking budgets.

This post confirms Coding parity with Sonnet 4 and Computer use gains Relative to Sonnet 4 Under These Scaffoldings. Users must replicate with their own format, tool sets, and thinking budgets before generalizing.

Key takeaways

• Haiku 4.5 offers Sonnet-4-level encoding performance at one-third the cost and more than twice the speed.
• It outperforms Sonnet 4 for computing tasks, improving responsiveness in Claude for Chrome and multi-agent flows in Claude Code.
• Recommended format: Use Sonnet 4.5 for multi-step planning and parallelization of execution with multiple workers in Haiku 4.5.
• The price is $1/$5 per million I/O codes; Available via Claude API, Amazon Bedrock, and Google Cloud Vertex AI.
• It was released under ASL-2 with a measured misalignment rate lower than the Sonnet 4.5 and Opus 4.1 in Anthropic tests.

Anthropic’s positioning of Claude Haiku 4.5 is strategically sound: through delivery Similar levels of programming performance to Cloud Sonet 4 in One third of the cost and More than twice the speedwhile Go beyond Sonnet 4 in computer usethe company gives developers a clean division between the planner and the executor—Sonnet 4.5 For multi-step planning and a combination of Haiku 4.5 workers for parallel execution – without forcing architectural changes (“Replace dropdown“Er API, Amazon Bedrock, Vertex AI). the ASL-2 Release, along with documentation Low misalignment rate from Sonnet 4.5 and Opus 4.1reduces friction in enterprise deployment where security gates and cost envelopes dominate deployment calculations.

verify Technical details, system card, model page and documentation . Feel free to check out our website GitHub page for tutorials, codes, and notebooks. Also, feel free to follow us on twitter Don’t forget to join us 100k+ mil SubReddit And subscribe to Our newsletter. I am waiting! Are you on telegram? Now you can join us on Telegram too.

Asif Razzaq is the CEO of Marktechpost Media Inc. As a visionary entrepreneur and engineer, Asif is committed to harnessing the potential of AI for social good. His most recent endeavor is the launch of the AI ​​media platform, Marktechpost, which features in-depth coverage of machine learning and deep learning news that is technically sound and easy to understand by a broad audience. The platform has more than 2 million views per month, which shows its popularity among the masses.