Anthropic has released Claude Opus 5.5, a lower-cost successor to Claude Opus 5 that leads Artificial Analysis’ Intelligence Index v4.3 and Vals AI’s Vals Index in overall intelligence benchmarks. Concurrently, TypeSafe, founded by OpenAI alumnus Diogo Almeida, introduced Jev, a general classification model designed to analyze text and return predefined outputs at lower cost than large language models.
- Claude Opus 5.5 supports up to 1 million input tokens and 128,000 output tokens, with reasoning levels ranging from low to max.
- The model scores first on Artificial Analysis’ Intelligence Index v4.3 (58) and Vals AI’s Vals Index (69.69 percent), outperforming GPT-6 Astra and Claude Fable 5.1.
- It features statistical watermarking, fast mode for 2.5x speed at 2x cost, and Zero Data Retention options for sensitive queries.
- Jev uses transformer-based architecture with reinforcement learning for calibrated decisions (RLCD) to classify text inputs into user-defined choices, scores, or yes/no answers.
- Claude Opus 5.5 is available via API and major cloud providers, while Jev is currently in early access.
Claude Opus 5.5 aims to improve performance on knowledge work tasks like writing business reports and research, addressing previous gaps against competitors. Jev provides a specialized tool for software systems requiring high-speed, low-cost decision-making data without the overhead of general-purpose language models.