Claude Sonnet 5

Proprietary dense model from Anthropic, the mid-tier workhorse of the Claude family. Parameter count is undisclosed. Native 1M-token context, knowledge cutoff 2025. Adaptive thinking is on by default and can be disabled at high effort or below; effort levels run low through max.

• Modalities: text, image, and PDF input; text output.
• Pricing: $2/1M input, $10/1M output.

Agents on Rails benchmark (Aug 2026, Le Mans round). 69.8% accuracy on 63 runs - the weakest Anthropic result recorded (the family ranges 92.1% down to 69.8% across the 16-model field). Notably, Sonnet 5 reaches for the right Rails API more often than Opus 4.8 (25.4% vs 15.9%) yet finishes six runs behind at twice the time (8m 11s vs 3m 36s median) - better recall does not buy the result. At $0.59 mean cost it is a cost-efficient Anthropic option, but the score is a reminder that recall is not accuracy.