Claude 3.5 Haiku
Language Retired Full comparison ↗ anthropic/claude-3-5-haiku · by Anthropic
· decoder-only-transformer
Pricing — 1 offering(s)
Input tokens
Retired — no longer sold · last price $0.80 / 1M tokens (input), until 2026-02-18
- $0.80 / 1M tokens (input) Historical 2024-10-22 → 2026-02-18
Output tokens
Retired — no longer sold · last price $4.00 / 1M tokens (output), until 2026-02-18
- $4.00 / 1M tokens (output) Historical 2024-10-22 → 2026-02-18
Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.
See how Claude 3.5 Haiku fits into a cost-aware routing setup
See how →Capability profile
How Claude 3.5 Haiku rates across core capability dimensions, with the task-level evidence behind each rating.
Punches above its weight for a fast model; better than most competing budget options.
Competitive at this tier; more reliable than GPT-4o mini on structured code tasks.
Strong function calling; widely deployed in agentic pipelines.
Excellent system prompt adherence — a Claude family hallmark.
200K context; the longest in its price tier at time of release.
Low latency; designed for high-throughput production workloads.
Operator guidance
Retired: Anthropic retired Claude 3.5 Haiku on 2026-02-19 and recommends Claude Haiku 4.5 ($1 / $5 per 1M tokens) as the replacement. Claude Haiku 5.5 is Anthropic's newest small model ($0.10 / $0.50 for prompts up to 100K tokens), and Claude Sonnet 5.5 ($2 / $10) is the step up for harder reasoning (Anthropic model deprecations and pricing pages, checked 2026-10-11).
Use cases
- Production chat and classification where latency matters
- Agentic pipelines needing reliable tool calling at moderate cost
- Long-document summarisation (200K window)
Limitations
- Superseded by Claude Haiku 4 in reasoning/coding; may not be the best long-term choice
- More expensive than GPT-4o mini or Llama 4 Maverick for simple workloads
- Capability ratings are qualitative, not a single cited benchmark run