Anthropic launched Claude Haiku 4.5, a latency-optimized “small” mannequin that delivers comparable ranges of coding efficiency to Claude Sonnet 4 whereas working greater than twice as quick at one-third the price. The mannequin is straight away out there by way of Anthropic’s API and in accomplice catalogs on Amazon Bedrock and Google Cloud Vertex AI. Pricing is $1/MTok enter and $5/MTok output. Anthropic positions Haiku 4.5 as a drop-in alternative for Haiku 3.5 and Sonnet 4 in cost-sensitive, interactive workloads.
Positioning and lineup
Haiku 4.5 targets real-time assistants, customer-support automations, and pair-programming the place tight latency budgets and throughput dominate. It surpasses Sonnet 4 on “pc use” duties—the GUI/browser manipulation underpinning merchandise like Claude for Chrome—and is described as materially bettering responsiveness in Claude Code for multi-agent tasks and fast prototyping. Anthropic makes clear that Sonnet 4.5 stays the frontier mannequin and “the most effective coding mannequin on the planet,” whereas Haiku 4.5 gives near-frontier efficiency with better cost-efficiency. A beneficial sample is Sonnet 4.5 for multi-step planning and parallel execution by a pool of Haiku 4.5 staff.
Availability, identifiers, and pricing
From day one, builders can name the mannequin (claude-haiku-4-5) on Anthropic’s API. Anthropic additionally states availability on Amazon Bedrock and Vertex AI; mannequin catalogs might replace area protection and IDs over time, however the firm confirms cloud availability within the launch publish. The API value for Haiku 4.5 is $1/MTok (enter) and $5/MTok (output), with prompt-caching listed at $1.25/MTok write and $0.10/MTok learn.
Benchmarks
Anthropic summarizes outcomes throughout normal and agentic suites and contains methodology particulars to qualify the numbers:
- SWE-bench Verified: easy scaffold with two instruments (bash, file edits), 73.3% averaged over 50 trials, no test-time compute, 128K considering finances, default sampling. Features a minor immediate addendum encouraging intensive instrument use and writing assessments first.
- Terminal-Bench: Terminus-2 agent, common over 11 runs (6 with out considering, 5 with 32K considering finances).
- OSWorld-Verified: 100 max steps, averaged throughout 4 runs with a 128K complete considering finances and 2K per-step configuration.
- AIME / MMMLU: averages over a number of runs utilizing default sampling and 128K considering budgets.
The publish emphasizes coding parity with Sonnet 4 and computer-use features relative to Sonnet 4 below these scaffolds. Customers ought to replicate with their very own orchestration, instrument stacks, and considering budgets earlier than generalizing.
Key Takeaways
- Haiku 4.5 delivers Sonnet-4-level coding efficiency at one-third the price and greater than twice the pace.
- It surpasses Sonnet 4 on computer-use duties, bettering responsiveness in Claude for Chrome and multi-agent flows in Claude Code.
- Beneficial orchestration: use Sonnet 4.5 for multi-step planning and parallelize execution with a number of Haiku 4.5 staff.
- Pricing is $1/$5 per million enter/output tokens; out there by way of Claude API, Amazon Bedrock, and Google Cloud Vertex AI.
- Launched below ASL-2 with a decrease measured misalignment fee than Sonnet 4.5 and Opus 4.1 in Anthropic’s assessments.
Anthropic’s positioning of Claude Haiku 4.5 is strategically sound: by delivering comparable ranges of coding efficiency to Claude Sonnet 4 at one-third the price and greater than twice the pace, whereas surpassing Sonnet 4 on pc use, the corporate offers devs a clear planner–executor break up—Sonnet 4.5 for multi-step planning and a pool of Haiku 4.5 staff for parallel execution—with out forcing architectural modifications (“drop-in alternative” throughout API, Amazon Bedrock, Vertex AI). The ASL-2 launch, coupled with a documented decrease misalignment fee than Sonnet 4.5 and Opus 4.1, lowers the friction for enterprise rollout the place security gates and price envelopes dominate deployment math.
Try the Technical particulars, system card, mannequin web page, and documentation . Be at liberty to take a look at our GitHub Web page for Tutorials, Codes and Notebooks. Additionally, be happy to observe us on Twitter and don’t overlook to hitch our 100k+ ML SubReddit and Subscribe to our Publication. Wait! are you on telegram? now you possibly can be a part of us on telegram as properly.
Asif Razzaq is the CEO of Marktechpost Media Inc.. As a visionary entrepreneur and engineer, Asif is dedicated to harnessing the potential of Synthetic Intelligence for social good. His most up-to-date endeavor is the launch of an Synthetic Intelligence Media Platform, Marktechpost, which stands out for its in-depth protection of machine studying and deep studying information that’s each technically sound and simply comprehensible by a large viewers. The platform boasts of over 2 million month-to-month views, illustrating its reputation amongst audiences.
