Claude Opus 5.5: 40% cheaper, frontier-grade performance

This week saw Claude Opus 5.5 released with a 40% cost reduction and comparable frontier performance to its predecessor. The new model achieves benchmark parity with Fable 5.1 at 40% lower cost and 30% faster output generation. One of the most significant cost savings is in cache reads, dropping from $0.50 to $0.20 per million tokens—a 60% reduction. This is particularly beneficial for agentic coding workflows, where the same system prompt, codebase context, or document corpus is frequently reused across many turns.

Vercel AI Gateway has added four new models: GLM-5.3 Flash, DeepSeek V4.1 Flash, Qwen 3.8 Flash, and Grok 4.7. These models provide 1M token context and multimodal support. The gateway abstracts away authentication, retry logic, and provider-level failover, allowing users to swap model strings without operational overhead.

Gemini 3.5 Transcribe is now available on AI Gateway, offering WebSocket-based live transcription in over 85 languages with custom vocabulary support and streaming output. This eliminates the need for a separate Google Speech-to-Text endpoint or credential set, reducing latency and operational surface area for real-time transcription tasks.

DeepSeek V4.1 Flash, Grok 4.7, and Qwen 3.8 Flash have also been launched on Vercel AI Gateway, offering features like 1M token context windows, vision support, and a configurable reasoning level parameter for tuning latency-depth tradeoffs. These models provide cost-competitive alternatives for coding agents and tool use, with low risk and easy integration options.