Anthropic’s new budget model gets much better at ignoring hidden commands
Anthropic released Claude Haiku 5.5, a cheaper model with stronger prompt-injection resistance than Haiku 4.5.
Anthropic released Claude Haiku 5.5, a lower-cost model for fast, repetitive work, available as claude-haiku-5-5 on the Claude Platform, AWS, Google Cloud, and Microsoft Azure. With cybersecurity safeguards off, it achieved arbitrary code execution in 4 of 410 Chrome V8 runs and completed 3.3% of multi-stage cyber challenges, far below Sonnet 5.5 (46.1%) and Opus 5.5 (67.6%). Production-style tests showed a 99.71% harmless response rate, Claude Code refusals of malicious requests at 84.3% (up from 66.6%), and stronger prompt-injection resistance than Haiku 4.5. Input and output prices are 90% lower for prompts up to 100,000 tokens, and Anthropic is also halving Sonnet 5.5 cache-read prices.