← All models

Claude Haiku 4.5

Language Available Full comparison ↗

anthropic/claude-haiku-4-5 · by Anthropic · decoder-only-transformer

Pricing — 1 offering(s)

Anthropic

Input tokens

  • $1.00 / 1M tokens (input) Current 2026-06-12 → present
    SCO-606 re-verification: price unchanged at $1/MTok. SCO-613 (2026-09-20): renamed slug/model.id/filename…

    SCO-606 re-verification: price unchanged at $1/MTok. SCO-613 (2026-09-20): renamed slug/model.id/filename from claude-haiku-4(-anthropic) to claude-haiku-4-5(-anthropic) — "Claude Haiku 4" (without ".5") never existed; Anthropic skipped the 4.0 point release for Haiku entirely (3.5 -> 4.5 directly). This entry's price was always correct (it was always tracking Haiku 4.5's real $1/$5 rate); only the name/slug/id were wrong. Companion renames in the same pass: modelglass-coding's registry/models/claude-haiku-4.yaml -> claude-haiku-4-5.yaml and this repo's own ontology/models/claude-haiku-4.yaml -> claude-haiku-4-5.yaml, plus closest_competitors citations updated in gpt-5-4-mini-openai.yaml, gpt-5-6-luna-openai.yaml, and claude-3-5-haiku-anthropic.yaml.

Output tokens

  • $5.00 / 1M tokens (output) Current 2026-06-12 → present

    SCO-606 re-verification: unchanged at $5/MTok. See the input tier's note for the SCO-613 rename history.

Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.

See how Claude Haiku 4.5 fits into a cost-aware routing setup

See how →

Capability profile

How Claude Haiku 4.5 rates across core capability dimensions, with the task-level evidence behind each rating.

reasoning Moderate
coding Strong

SWE-bench Verified (500 human-verified GitHub issues; patch must pass the repo's own tests; no file-location hints): resolves 73.3% — fails roughly 1 issue in 4. Vendor-reported (Anthropic, Claude Haiku 4.5 launch; agentless-style two-tool scaffold). Full harness/version detail: companion modelglass-coding entry.

tool use Strong
instruction following Strong
context window Strong

200K-token context.

multilingual Moderate
speed Strong

Smallest/fastest Claude 4 tier — low latency and cost.

cost efficiency Strong

Ratings are estimated — limited independent data is available for this model.

Coding benchmarks

View the full coding benchmark leaderboard →
Benchmark Score Harness Source
swe-bench-verified ↗ 73.3% 2025-10 agentless 2.0 🏢 vendor ↗
terminal-bench-2-1 44.2% 2026-08 terminus-2 🔬 independent ↗

Operator guidance

Route here for throughput and cost. Step up to Sonnet 4 when a task needs more reasoning or coding depth than Haiku reliably provides.

Use cases

Limitations

Citations