AI developer tools · What shipped

Claude Opus 5 was launched on July 24, 2026, as a thoughtful and proactive model that approaches the frontier intelligence of Claude Fable 5 at half the price. Claude Sonnet 5 was launched on June 30, 2026, featuring substantial improvements over Sonnet 4.6 in reasoning, tool use, coding, and knowledge work with agentic capabilities. Claude Opus 4.8 launched on May 28, 2026, showing improvements in coding, agentic skills, reasoning, and practical knowledge work tasks over Opus 4.7. Claude Opus 4.7 launched on April 16, 2026, with improvements in software engineering, complex long-running coding tasks, and higher-resolution vision capabilities. Claude Sonnet 4.6 launched on February 17, 2026, featuring full upgrades across coding, computer use, long-context reasoning, agent planning, and design work, with a 1M token context window in beta. Claude Opus 4.6 launched on February 5, 2026, with improved coding skills. Claude Opus 4.5 launched on November 24, 2025, as the most powerful frontier model to date. Claude Sonnet 4.5 launched on September 29, 2025, as the best model for real-world agents, coding, and computer use. Claude Haiku 4.5 launched on October 15, 2025, as the fastest and most cost-efficient model, matching Sonnet 4's performance on coding and agent tasks.

Ornith-1.5, a major open-source AI model release, introduces three scaled variants (397B MoE, 35B MoE, and 9B Dense) with a self-improvement framework that continuously generates new training tasks, constructs task-specific scaffolds, and produces solution rollouts for reinforcement learning. The flagship Ornith-1.5-397B achieves 86.1 on Terminal-Bench 2.1 and 56.0 on DeepSWE, matching Claude Opus 4.8 performance while outperforming comparable open-source models including GLM-5.2 and DeepSeek-V4-Flash-0731; the 35B variant significantly outperforms Qwen 3.6-35B across coding and agentic benchmarks, and the compact 9B model achieves 47.0 on Terminal-Bench 2.1 and 70.6 on SWE-Bench Verified while exceeding performance of much larger models like Gemma 4-31B.

Gemma 4 is now up to 90% faster in Ollama 0.31 on Apple Silicon via multi-token prediction, as measured using the Aider polyglot benchmark. Ollama 0.30 added improved performance and GGUF model compatibility through llama.cpp, expanding support across more hardware. NVIDIA Nemotron 3.5 Lightning, a 30 billion parameter open model built for agents, is now available on Ollama, while Meta's Muse Glimmer—a 30B multimodal model under Apache 2.0 license—is available with native DFlash and image input support accelerated by Ollama's MLX engine.

CUDA Python 1.0 has been released with stable APIs and semantic versioning commitments across multiple components. The release includes cuda.core 1.0.0 for CUDA runtime access, cuda.compute 1.0.0 for parallel algorithms, cuda.bindings 13.3.0 for low-level C API bindings, nvmath-python 1.0 for math libraries, and cuda-pathfinder for locating CUDA components. Key new capabilities in cuda.core include green contexts for GPU streaming multiprocessor partitioning, process checkpointing, and inter-process GPU memory sharing. The release establishes Python as an officially supported first-class citizen on the CUDA platform alongside C++, with NVIDIA committing to feature-complete parity and predictable deprecation policies. The unified foundation allows Python GPU libraries to interoperate without private binding layers, with existing adopters including CuPy, PyTorch, and NVIDIA's communications and math libraries.