AI Model Timeline

Tracking the accelerating release frequency of frontier AI models.

Claude Haiku 5.5

Anthropic·

LLMssmallcloud

Overview

Haiku-tier successor to Claude Haiku 4.5 and the third Claude 5.5 model, aimed at high-volume work such as summaries, compaction, classification and coding subagents. Priced at $0.10 / $0.50 per million input/output tokens for prompts up to 100K tokens and $0.50 / $2.50 above that (Haiku 4.5: $1 / $5); Anthropic puts the average running cost about 75% below Haiku 4.5. 1M-token context, 128K max output. First Haiku with an adjustable effort setting (low to max). Cybersecurity safeguards stricter than Haiku 4.5, looser than Sonnet 5.5; biology safeguards as on Sonnet 5.5. Anthropic reports no GPQA, SWE-bench Pro or MMLU-Pro figure.

Capabilities and innovations

Agentic coding (subagent)Computer useExtended thinking with effort controlTool useVision input1M token context windowAPI: claude-haiku-5-5First Haiku with effort control (low to max)Tiered pricing at the 100K-token prompt boundary

Benchmarks

BenchmarkScoreSource
Humanity’s Last Exam (no tools)45.9%vendor
HLE 45.9% no tools (57.4% with tools), Anthropic launch table. Artificial Analysis measures 44.4% (Max effort), within 3 points, so the vendor value stands. The same table restates Haiku 4.5 as 10.2% no tools / 18.7% with tools.
Terminal-Bench 4.035.35%independent
Terminal-Bench 4.0 35.35% (± 2.20), measured by Vals AI at max effort, rank 12 of 45 (accessed 2026-10-12). 4.0 is its own generation and not comparable with the 2.x or 3.0 fields.
  • GPQA Diamond: independent evaluation pending — Not in the launch table; Artificial Analysis lists no GPQA value for Haiku 5.5 (checked 2026-10-12).
  • SWE-bench Pro: not reported by the vendor — Not in the launch table; Anthropic reports FrontierCode 1.1 and Terminal-Bench 4.0 for coding instead.
  • MMLU-Pro: independent evaluation pending — Not in the launch table; no Vals AI or Artificial Analysis value yet (checked 2026-10-12).
  • MMMU-Pro: independent evaluation pending — Not in the launch table; Artificial Analysis publishes no MMMU-Pro value yet (checked 2026-10-12).
  • Terminal-Bench 2.x: not reported by the vendor — No 2.x figure; 4.0 is the generation Anthropic and Vals AI report.
  • Terminal-Bench 3.0: not reported by the vendor — Not in the launch table; 4.0 is the current release.
  • ARC-AGI-2: independent evaluation pending — No ARC Prize result yet (checked 2026-10-12).

API pricing

$0.1 per million input tokens, $0.5 per million output tokens (USD, cheapest OpenRouter endpoint, checked 2026-10-12)

Links

More from Anthropic

Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.