Claude Haiku 4.5
Anthropic's speed-and-cost tier, released 15 October 2025 — Sonnet 4-level coding at a third of the cost and more than twice the speed.
Claude Haiku 4.5 specs at a glance
| Maker | Anthropic |
| Released | 15 October 2025 |
| Context window | 200,000 tokens |
| Max output | Up to 64,000 tokens |
| API pricing | $1 / $5 per 1M input / output tokens |
| Reasoning | Extended thinking (first in the Haiku line) |
| Tools | Coding, bash, web search, computer use |
| Best at | High-volume, latency-sensitive, cost-sensitive tasks; subagents |
Calculate API costs for this model → Compare its monthly bill against 15 other rate cards.
Where Haiku 4.5 fits in mid-2026
Haiku 4.5 launched in October 2025 with a striking pitch: Sonnet 4-level coding performance at one-third the cost and more than twice the speed. It also brought extended thinking to the Haiku line for the first time, with controllable reasoning depth and full tool support.
Nine months on, it remains the floor of Anthropic's lineup — the model you route to when the task is simple, the volume is high, or latency matters more than depth. Its most common 2026 role is as the subagent workhorse inside multi-agent systems: a frontier model plans, and a fleet of Haikus executes.
The trade-off is the 200K context window — one-fifth of what Sonnet 5 and Opus 4.8 offer — and a 64K output cap.
Pricing
Haiku 4.5 costs $1 per 1M input tokens and $5 per 1M output tokens. That is one-fifth of Opus 4.8's rate card, and half of Sonnet 5's launch-discount pricing. For classification, extraction, routing and other short-output tasks, it is the cheapest way to stay inside the Claude ecosystem.
Who should use Claude Haiku 4.5
- High-volume pipelines — classification, extraction, summarization at scale.
- Latency-sensitive products — chat, autocomplete, real-time assistants.
- Multi-agent architectures — cheap, fast subagents under a frontier-model orchestrator.
- Budget-capped coding — Sonnet 4-class coding without Sonnet pricing.
How Claude Haiku 4.5 compares
- Claude Sonnet 5 — the next tier up →
- Claude Opus 4.8 — the Opus flagship →
- DeepSeek V4 — the open-weight budget alternative →
- Kimi K2.6 — open-weight coding on a budget →
Frequently asked questions
When was Claude Haiku 4.5 released?
Anthropic released Claude Haiku 4.5 on 15 October 2025. As of mid-2026 it remains Anthropic's fastest and cheapest model.
How much does Claude Haiku 4.5 cost?
$1 per million input tokens and $5 per million output tokens — one-fifth the price of Claude Opus 4.8.
What is Claude Haiku 4.5's context window?
200,000 tokens, with up to 64,000 output tokens — smaller than the 1M windows of Sonnet 5 and Opus 4.8.
How good is it at coding?
Anthropic reports similar coding performance to Claude Sonnet 4 at one-third the cost and more than twice the speed, with full coding, bash, web-search and computer-use tool support.
Does it support extended thinking?
Yes — Haiku 4.5 was the first Haiku-line model with extended thinking, enabling controllable reasoning depth and tool-assisted workflows.