Claude Haiku 5.5 vs Sonnet 5.5 vs Opus 5.5: which to use

Quick answer: Anthropic's own guidance is to start with Claude Opus 5.5 for most workloads. Use Sonnet 5.5 when you want speed and capability for everyday coding and agents, and Haiku 5.5 for the lowest latency and price, such as classification, extraction, routing and sub-agents. All three have a 1M-token context window and 128K max output. Standard API prices per million input/output tokens: Opus 5.5 $4/$20, Sonnet 5.5 $2/$10, Haiku 5.5 from $0.10/$0.50. For the hardest long-horizon work, Anthropic points to Claude Fable 5.1 ($10/$50).

Key facts

• Release dates: Opus 5.5 on 22 Sep 2026, Sonnet 5.5 on 28 Sep, Haiku 5.5 on 7 Oct.
• API IDs: claude-opus-5-5, claude-sonnet-5-5, claude-haiku-5-5.
• Relative speed (Anthropic): Opus moderate, Sonnet fast, Haiku fastest.
• Default effort: Opus 5.5 medium, Sonnet 5.5 high, Haiku 5.5 medium.
• Cache reads: $0.20 (Opus), $0.10 (Sonnet), $0.01 (Haiku, prompts up to 100K) per million tokens.
• Batch API: 50% off on all of them.

Side by side

• Claude Opus 5.5: "for long-running agentic coding and knowledge work". $4 input / $20 output per MTok. Adaptive thinking is always on and can't be disabled. Fast mode (research preview) costs $8/$40.
• Claude Sonnet 5.5: "the best combination of speed and intelligence". $2 / $10. Adaptive thinking. Cache reads dropped to $0.10 on 7 October.
• Claude Haiku 5.5: for "high-volume, latency-sensitive tasks such as classification, extraction, and routing". $0.10 / $0.50 for prompts up to 100K tokens, and $0.50 / $2.50 above that.

Descriptions and prices are from Anthropic's models overview and pricing page. Haiku 5.5 is the only one priced by prompt length. Per the pricing page, the other current models bill the full 1M context at standard rates.

Anthropic's model selection matrix

• Opus 5.5, for complex agentic coding and enterprise work: multihour autonomous coding agents, large-scale refactoring, complex systems engineering, vision-heavy workflows, computer use.
• Sonnet 5.5, for everyday coding, agent and enterprise workloads: code generation, data analysis, content creation, visual understanding, agentic tool use.
• Haiku 5.5, for the lowest latency and price: real-time apps, high-volume processing, cost-sensitive deployments, sub-agent tasks.
• Fable 5.1, for the highest capability: agent sessions that run for hours, multistep deep research, analysis carried through to a finished document.

The choosing-a-model guide describes two ways to start. Efficiency-first means starting on Haiku 5.5 and upgrading only where it falls short. Capability-first means starting on Opus 5.5 and moving down, or lowering effort, once things work. Anthropic notes that tuning effort is often a better lever than switching models.

In the Claude apps and Claude Code

On claude.com/pricing, Sonnet and Haiku are on every plan, including Free, while Opus needs Pro or above. Fable needs usage credits on Pro and is capped at 50% of weekly limits on Max. In Claude Code, version 2.1.293 made Haiku 5.5 the default Haiku model. If you're moving API code from older models, read Anthropic's migration notes first. Opus 5.5 rejects disabled thinking and forced tool use, and Sonnet 5.5 and Haiku 5.5 each have their own breaking changes, listed in the release notes.