[AINews] Claude Haiku 5.5 — better than GPT-6 Luna at the same pricing
Haiku 5.5 is framed as Anthropic’s cheap worker model, matching Luna’s sticker price while beating it on several cited evals.
Latent Space says Anthropic shipped Haiku’s first major update in about a year, adding effort controls, adaptive thinking, and a 1M-token context. The piece cites Artificial Analysis scoring Haiku 5.5 at 43 on its Intelligence Index, ahead of GPT-6 Luna’s 38, while noting Haiku uses more tokens at higher effort. The pricing is tiered, with cheaper rates under 100K-token prompts and a 5x step above that threshold. The story’s caveats are mostly about real task cost, weaker factual recall versus some rivals, and an over-refusal bug affecting automation results. Latent Space's note
Latent Space says Anthropic shipped Haiku’s first major update in about a year, adding effort controls, adaptive thinking, and a 1M-token context. The piece cites Artificial Analysis scoring Haiku 5.5 at 43 on its Intelligence Index, ahead of GPT-6 Luna’s 38, while noting Haiku uses more tokens at higher effort. The pricing is tiered, with cheaper rates under 100K-token prompts and a 5x step above that threshold. The story’s caveats are mostly about real task cost, weaker factual recall versus some rivals, and an over-refusal bug affecting automation results. Latent Space's note
score 8