Overview
- Anthropic made Sonnet 5 the default reasoning tier and Vercel listed it on the AI Gateway with launch pricing at $2/$10 per million input/output tokens that increases after August 31.
- GitLab’s Duo Agent benchmarks show Sonnet 5 resolved 8.8% more issues than Sonnet 4.6 and completed multi-step workflows that previously stalled, a result echoed by early testers.
- Independent comparisons show Sonnet 5 matches Opus 4.8 on coding and agentic tasks, allowing many teams to drop expensive Opus routing and capture a roughly 50–67% cost cut during the launch price period.
- Developer friction is falling as GitHub Copilot ships natively in JetBrains IDEs and Google’s Gemini 3.1 Flash Lite Image appears on Vercel at about $0.034 per 1,000 images with single-call text+image output and sub-4-second latency.
- ADK for Go 2.0 introduces a declarative graph API with durable state and human-in-the-loop gates, which simplifies long-running multi-agent orchestration and reduces brittle imperative code in production pipelines.