Jonathon's AI Wiki

Claude Haiku 5.5

3 min read

Source: ai-research/claude-code-docs-changelog-2026-10-09.md (code.claude.com changelog, v2.1.293, October 7, 2026). Product: Claude — Haiku tier URL: https://code.claude.com/docs/en/changelog

Claude Haiku 5.5 (claude-haiku-5-5) is the new default Haiku model on the Anthropic API, released on October 7, 2026 with Claude Code v2.1.293. It is the third and final model in the Claude 5.5 family, completing the lineup started with Opus 5.5 (Week 39) and Sonnet 5.5 (Week 40).

Key Takeaways

  • Model ID: claude-haiku-5-5
  • Pricing: 0.50 output per Mtok (the lowest-cost model in the Claude 5.5 family)
  • Context window: 1M tokens
  • Default Haiku on the Anthropic API from Claude Code v2.1.293 onward
  • Completes the Claude 5.5 family: Opus 5.5 (20), Sonnet 5.5 (10), Haiku 5.5 (0.50)
  • Use as the fast, cheap tier for high-throughput tasks: tool routing, classification, quick drafts, bulk processing

Pricing

MetricRate
Input$0.10/Mtok
Output$0.50/Mtok
Context window1M tokens

For comparison:

ModelInputOutputContext
Claude Haiku 5.5$0.10$0.501M
Claude Sonnet 5.5$2.00$10.001M
Claude Opus 5.5$4.00$20.001M

The Claude 5.5 Family

Haiku 5.5 is the budget tier of the family launched across Weeks 39–41:

  • Claude Opus 5.5 — claude-opus-5-5, 20/Mtok, default Opus on Anthropic API since v2.1.280 (Week 39)
  • Claude Sonnet 5.5 — claude-sonnet-5-5, 10/Mtok (cache reads $0.10/Mtok from v2.1.296), default Sonnet on Anthropic API since v2.1.284 (Week 40)
  • Claude Haiku 5.5 — claude-haiku-5-5, 0.50/Mtok, default Haiku on Anthropic API since v2.1.293 (Week 41)

Common Use Cases

  • High-volume classification and routing — cheapest way to label, filter or route at scale
  • Workflow subagents (CLAUDE_CODE_WORKFLOW_SUBAGENT_MODEL=claude-haiku-5-5) for steps that do not need deep reasoning
  • Tool selection and dispatch in multi-step pipelines where each step is well-defined
  • Draft generation before a Sonnet or Opus pass refines the output
  • Batch annotation of datasets where speed and cost matter more than nuance

Claude Code Integration

In Claude Code v2.1.293+:

  • Default Haiku: Claude Code selects claude-haiku-5-5 wherever the Haiku model family is the target (e.g., the /model picker’s Haiku row).
  • ANTHROPIC_DEFAULT_HAIKU_MODEL=claude-haiku-5-5 pins this as the session Haiku.
  • effort parameter on Agent tool (v2.1.292): spawn Haiku 5.5 subagents at medium effort for bulk, low-cost parallel steps.
  • CLAUDE_CODE_WORKFLOW_SUBAGENT_MODEL=claude-haiku-5-5 keeps workflow subagents on the cheap tier when the session runs Opus.

Open Questions

  • Bedrock, Vertex and Foundry defaults: The v2.1.293 changelog note says “the Anthropic API.” Whether the default Haiku on those providers also updated simultaneously is not stated.
  • Cache reads pricing: The v2.1.296 changelog explicitly cut Sonnet 5.5 cache reads from 0.10. The Haiku 5.5 cache read price is not stated in the available snapshot.
  • Performance benchmarks: No benchmark figures are included in the changelog entry. External benchmark data is not yet available (as of the ingest date).

Graph View

Backlinks