Anthropic has launched Claude Opus 5.5, reducing its pricing by 20%. The model costs $4 per million input tokens and $20 per million output tokens. Cache reads are priced at $0.20 per million tokens, representing a 60% reduction. A high-performance mode is available for $8 per million input tokens and $40 per million output tokens.
Performance metrics for Opus 5.5 show a 30% increase in generation speed compared to Opus 5. On the Terminal-Bench 4.0 coding test, the model achieved a score of 66.4%, while AutomationBench task completion reached 40%. In scientific research benchmarks, the model scored 58.7% on Terminal-Bench-Science 0.1, compared to 29% for its predecessor.
OpenAI responded by releasing two GPT-6 models, Sol and Luna, priced at 50% less than their predecessors. Sol costs $2 per million input tokens and $10 per million output tokens, while Luna is priced at $0.10 and $0.50 respectively. OpenAI also implemented a 90% discount on cached input token reads for these models.
Sol is designed for recurring coding and agent tasks, achieving 68.8% on the DeepSWE v1.1 test. Luna targets high-volume routine operations like summarization. Both companies emphasized improved safety and reduced verbosity in their latest releases, with Anthropic noting an 85% decrease in successful attempts to circumvent containment boundaries compared to Opus 5.
