Anthropic Claude Haiku 4.5 Release: Near‑Frontier Coding Performance at One‑Third Cost and Double Speed

TL;DR

Claude Haiku 4.5 is now generally available and offers coding performance comparable to the frontier Claude Sonnet 4 model while costing about one‑third as much and running more than twice as fast, making it ideal for low‑latency, cost‑sensitive applications.

Near‑Frontier Performance at Small Model Scale

Claude Haiku 4.5 matches the coding quality of Claude Sonnet 4, which was state‑of‑the‑art five months ago. In the Augment agentic coding benchmark it reaches 90 % of Sonnet 4.5’s performance, and it even surpasses Sonnet 4 on specific tasks such as computer‑use interactions. This demonstrates that a small model can now deliver near‑frontier results without the typical trade‑off between quality and efficiency.

Cost and Speed Advantages

  • Cost: Pricing is $1 per million input tokens and $5 per million output tokens, roughly one‑third the cost of Sonnet 4.
  • Speed: Haiku 4.5 runs 4–5× faster than Sonnet 4.5, delivering responses in real‑time for chat assistants, customer‑service bots, and pair‑programming tools.
  • Throughput: The model’s efficiency expands the amount of work that can be done within existing usage limits, enabling more responsive AI‑assisted development workflows.

Real‑World Impact on Applications

  • Claude for Chrome: Faster inference makes the browser extension more useful for on‑the‑fly assistance.
  • Claude Code & Warp: Developers experience markedly lower latency when iterating on code, especially in multi‑agent or rapid‑prototyping scenarios.
  • Agentic Orchestration: Sonnet 4.5 can decompose complex problems into multi‑step plans and delegate subtasks to multiple Haiku 4.5 instances, enabling parallel execution and higher overall throughput.

Benchmark Highlights

  • SWE‑bench Verified: Haiku 4.5 achieves coding performance comparable to frontier models while operating at a fraction of the cost.
  • Instruction‑following (slide text generation): 65 % accuracy versus 44 % for Anthropic’s premium tier model, indicating strong general‑purpose instruction compliance.
  • GitHub Copilot integration: Early tests show code generation quality on par with Sonnet 4 but with noticeably faster response times.

"Claude Haiku 4.5 hit a sweet spot we didn't think was possible: near‑frontier coding quality with blazing speed and cost efficiency." – Anthropic release statement

"Claude Haiku 4.5 is a leap forward for agentic coding, particularly for sub‑agent orchestration and computer use tasks." – Anthropic release statement

Safety and Alignment

  • Alignment: Automated assessments show a statistically significant lower rate of misaligned behaviors compared to Claude Haiku 3.5, Sonnet 4.5, and Opus 4.1, making Haiku 4.5 Anthropic’s safest model to date.
  • Risk Profile: Limited CBRN weapon‑production risk; classified under AI Safety Level 2 (ASL‑2), less restrictive than the ASL‑3 rating of Sonnet 4.5 and Opus 4.1.
  • System Card: Full safety methodology and classification rationale are documented in the Claude Haiku 4.5 system card.

Availability and Integration

  • API Access: Developers can call the model via the Claude API using the identifier claude-haiku-4-5.
  • Cloud Platforms: The model is available on Amazon Bedrock and Google Cloud Vertex AI as a drop‑in replacement for Haiku 3.5 and Sonnet 4 at the most economical price point.
  • Anthropic Products: Included in Claude Code, Claude for Chrome, and other Anthropic apps.

Future Directions

Claude Haiku 4.5 illustrates a shift toward models that combine high reasoning ability with real‑time responsiveness. Anthropic suggests that this class of models will enable new use cases where latency and cost have previously been prohibitive, such as large‑scale sub‑agent orchestration and interactive AI assistants.

References

Sources

Related