Anthropic Launches Claude Haiku 4.5

Anthropic launched Claude Haiku 4.5 today (October 15, 2025), delivering performance that matches its flagship Sonnet 4 model on coding, computer use, and agent-based tasks at one-third the cost and more than twice the speed.

The new model scored 73.3% on SWE-bench Verified, a benchmark that measures AI coding ability by tasking models with fixing real-world bugs in open-source Python repositories using only the original issue description. Solutions are judged by whether they pass each repository’s existing unit tests, with no partial credit.

The score places Haiku 4.5 among the world’s top coding models, equaling Sonnet 4’s performance while dramatically undercutting it on price and latency.

Anthropic priced Haiku 4.5 at $1 per million input tokens and $5 per million output tokens. Developers can push costs even lower through prompt caching, which cuts repeated input costs by up to 90%, and the Message Batches API, which offers a 50% discount for workloads that tolerate a 24-hour processing window. Combined, the two features can reduce input token costs by 95% for cached prompts processed in batch mode.

The model surpasses Sonnet 4 on computer use tasks—the GUI and browser manipulation capabilities that power Claude for Chrome, Anthropic’s browser-based agent extension launched in August. Computer use allows Claude to read web pages, click buttons, and fill forms directly within Chrome, a feature that Haiku 4.5 now handles faster than its larger sibling. The extension is currently available to 1,000 subscribers on Anthropic’s Max plan, with a broader waitlist open.

Anthropic positioned Haiku 4.5 as a drop-in replacement for both Haiku 3.5 and Sonnet 4 in cost-sensitive, interactive workloads. The company recommends using Sonnet 4.5—still Anthropic’s frontier model—for multi-step planning while deploying parallel execution by a pool of Haiku 4.5 workers for cost and throughput efficiency. Sonnet 4.5 remains “the best coding model in the world,” according to Anthropic.

The Haiku family debuted in March 2024 as part of the Claude 3 series, which introduced tiered models—Haiku, Sonnet, and Opus—balancing speed, cost, and intelligence. Claude 3.5 Haiku followed in October 2024 with an 8,192-token maximum output and training data through July 2024. The 4.5 release continues Anthropic’s pattern of shipping smaller models that punch above their weight class on targeted tasks.

Availability is immediate across Claude.ai on web, iOS, and Android. Developers can access Haiku 4.5 through Anthropic’s API, Amazon Bedrock, and Google Cloud Vertex AI. Target use cases include real-time assistants, customer-support automation, and pair-programming environments where latency and throughput matter. The model also improves responsiveness in Claude Code, Anthropic’s multi-agent code editor, for rapid prototyping workflows.

The launch arrives as Anthropic competes in the “AI browser wars.” Perplexity shipped its Comet browser in July 2025, OpenAI is reportedly building its own AI browser, and Google has integrated Gemini with Chrome. Haiku 4.5’s speed advantage on computer use tasks positions Anthropic to scale browser automation without the latency penalties of larger models.

Anthropic’s pricing and performance gains reflect broader industry pressure to deliver capable models at inference costs low enough for production deployment. With Haiku 4.5, developers can now run near-frontier coding and automation workloads at a fraction of the cost, potentially shifting economic feasibility for agent-based applications that were previously too expensive to scale.