Anthropic Launches Claude Sonnet 5
Anthropic released Claude Sonnet 5 on June 30, 2026, describing it as the most agentic Sonnet model to date. The model is built to plan multi-step work, operate tools like browsers and terminals, and run autonomously for longer stretches than previous Sonnet generations — capabilities that until recently required stepping up to a larger, more expensive model tier. Anthropic reports that Sonnet 5's agentic performance now approaches Opus 4.8 on reasoning, tool use, coding, and knowledge-work benchmarks, while keeping Sonnet-tier pricing.
Sonnet 5 is available immediately across Anthropic's product surface: it's the default model for Free and Pro plan users, and it's accessible to Max, Team, and Enterprise customers, as well as through Claude Code and the Claude Platform. Developers can call it via the API using the model identifier claude-sonnet-5. Pricing is introductory through August 31, 2026, at $2 per million input tokens and $10 per million output tokens, stepping up to standard pricing of $3/$15 per million tokens afterward.
For teams building agentic workflows, the practical upshot is that tasks previously requiring escalation to Opus-tier models — longer tool-use chains, multi-file coding tasks, or research tasks needing several rounds of self-correction — may now be handled by the cheaper Sonnet tier. Anthropic also reports lower rates of undesirable agentic behavior in safety evaluations, though cybersecurity-relevant capabilities remain deliberately more restricted than in Opus models.
Read more — Anthropic
IntelliJ IDEA 2026.1.4 Ships Bug Fixes for Git, Docker, and Gradle-on-WSL
JetBrains released IntelliJ IDEA 2026.1.4, a maintenance update focused entirely on stability fixes rather than new features. The active Git branch indicator now updates correctly after branch switches — a regression that had been showing stale branch names in the status bar. On the container tooling side, a bug that prevented PHP interpreter creation when a Docker Compose file specified a pull_policy value has been resolved, unblocking PHP developers using Compose-based dev environments.
Two additional fixes target hybrid Windows/Linux workflows: Gradle sync no longer reports false failure indicators when running Gradle 9.5.0 under WSL, and Dev Container connections that were failing with an "Unknown Docker endpoint schema" error now connect correctly. Teams that had worked around the WSL Gradle sync issue by ignoring red status indicators can drop that workaround.
The update is available through the IDE's built-in updater, the JetBrains Toolbox App, Ubuntu snaps, or a direct download from the JetBrains website.
Read more — The JetBrains Blog
Gemma 4 Gets Up to 3x Faster Inference on Apple Silicon via MLX
Google's Gemma 4 model family is now significantly faster on Apple Silicon, thanks to multi-token prediction (MTP) drafters running through the MLX framework, as detailed in a joint Ollama/Hugging Face rollout. MTP works via speculative decoding: a small drafter model proposes several future tokens at once, and the full Gemma 4 model verifies them in a single forward pass, producing identical output to standard decoding but substantially faster — Google reports up to a 3x speedup with no quality or reasoning degradation.
On Apple Silicon specifically, Ollama 0.31 now uses MLX to serve these MTP-accelerated Gemma 4 checkpoints, and early benchmarks (measured with the Aider polyglot benchmark) show up to 90% faster completion times when Gemma 4 is used as the backing model for coding agents. MLX's advantage over Ollama's default GGUF-based path comes from talking directly to the Metal runtime rather than going through GGUF quantization overhead, which independently accounts for another 10-20% speedup even without MTP.
Model weights, including MLX-quantized variants, are available on Hugging Face in the Gemma 4 and mlx-community collections, and are compatible with MLX, transformers, vLLM, SGLang, and Ollama. For developers running local coding agents on Mac hardware, this closes a meaningful chunk of the latency gap between local open-weight models and hosted frontier APIs.
Read more — Ollama Blog
