FULL STORY

Claude Haiku 5.5: From Leak to Launch and Benchmarks

Hours after leaking in a Claude Code update, Anthropic launched Claude Haiku 5.5 as its cheapest, fastest small model with ~75% lower costs. Early benchmarks rank it second in intelligence but flag surprisingly high token usage.

2026-10-08 ~ 2026-10-08 · 3 episodes · 42 posts

Episode 1 · Claude Haiku 5.5 Leaked in Claude Code, Priced Against GPT-6 Luna (2026-10-08, 3 posts)

Claude Haiku 5.5 has surfaced in Claude Code ahead of launch, with leaked pricing matching GPT-6 Luna at $0.10/M input tokens under 100K context, and unverified benchmarks claiming it beats Luna across the board.

Episode 2 · Anthropic Launches Claude Haiku 5.5 with ~75% Lower Cost (2026-10-08, 35 posts)

On October 8, Anthropic officially released Claude Haiku 5.5, calling it its "cheapest, fastest, and most capable" small model to date, with average running costs about 75% lower than the previous-generation Haiku 4.5; Sonnet cache pricing was also cut at the same time. The news came directly from Anthropic's official accounts @claudeai and @AnthropicAI, making it a confirmed official release.

Confirmed

  • Claude Haiku 5.5 has officially launched, with average running costs about 75% lower than Claude Haiku 4.5.
  • It is positioned as a high-value small model focused on low-latency, low-cost scenarios for high-volume, cost-sensitive tasks.
  • Anthropic says it delivers significant improvements over Haiku 4.5 in three areas: coding, computer use, and knowledge work.
  • Sonnet cache prices have been reduced in tandem.

Why it matters

  • A 75% cost cut dramatically lowers the barrier to using small models, directly influencing model selection for high-traffic, cost-sensitive applications such as customer service, batch processing, and agent tasks.
  • Improvements in coding and computer use mean small models can take on more tasks that previously required large models, intensifying price-performance competition in the small-model space.

15 more related posts →

Episode 3 · Claude Haiku 5.5 Ranks Second on Intelligence Index but Burns Tokens (2026-10-08, 4 posts)

Artificial Analysis's evaluation shows Claude Haiku 5.5 (max) scoring an intelligence index of 43, ranking 2nd among 179 models, and reaching 1578 Elo on AA-Briefcase. However, it consumes about 162k output tokens per task, roughly triple that of GPT-6 Luna.