Claude Haiku 5.5

Anthropic's new small model, released 7 October 2026. The headline is efficiency: about 75% cheaper to run than Haiku 4.5, and far more capable.

Links

• Announcement: https://www.anthropic.com/claude-haiku-5-5 (7 Oct 2026)
• Theo's reaction: https://x.com/theo/status/2108741275503763548 (10 Oct 2026)

Key points

• Price: $0.10 per million input tokens and $0.50 per million output tokens (prompts up to 100K), against $1.00 and $5.00 for Haiku 4.5. Anthropic says about 75% cheaper to run on average.
• Speed: Anthropic's fastest model at standard speed. One customer reports up to 2.5x faster inference per agent turn.
• Capability jump over Haiku 4.5: OSWorld 2.1 (offline subset) 72.4% vs 15.7%; Terminal-Bench 4.0 39.2% vs 0.0%; Humanity's Last Exam (no tools) 45.9% vs 10.2%; GDPval-AA v2.1 1620 vs 735 Elo.
• Effort setting: adjustable, so users can trade cost against capability.
• Availability: all platforms including AWS, Google Cloud and Azure. Model ID claude-haiku-5-5.

Testimony: Theo

Developer and YouTuber Theo (t3.gg) had written in September that "Anthropic has no small models that are worth using right now." On 10 October he replied: "This is no longer true. Haiku has placed Anthropic in a position of complete and utter domination of the market." (143K views at time of capture.) One developer's reaction, not a benchmark, but a useful signal that the cheap tier is now competitive.

Why it matters

Cheaper, capable small models make agent workloads that were too expensive to run at volume practical. The gain here is from efficiency, not a bigger frontier model.