AI developer tools · What shipped

Gemini 3.7 Flash released as generally available for coding and agentic tasks with substantial improvements in software engineering and web development. Gemini 3.6 Flash and 3.5 Flash-Lite also reached GA status, with 3.6 Flash offering improved token efficiency and reduced output verbosity at lower cost than 3.5 Flash, while 3.5 Flash-Lite provides low-latency, cost-effective automation for high-volume use cases. Gemini Omni Flash reached GA with new video generation capabilities including video extension, interpolation between images, and resolution control up to 4K. Gemini 3.5 Transcribe models launched in GA with two dedicated speech-to-text variants: the standard model supporting 85+ languages with speaker diarization and custom vocabulary biasing, and a live streaming version using WebSockets for bidirectional real-time transcription. Robotics ER 2 models entered public preview with advanced spatial reasoning and multi-robot coordination capabilities optimized for real-time streaming. Agentic video understanding released for Gemini 3.7 Flash and earlier Flash models, reducing token usage by up to 88% for long-form video content through dynamic timeline navigation.

Claude Fable 5.1 and Claude Mythos 5.1 have been released as the latest AI coding and knowledge work models from Anthropic. Fable 5.1 achieves significantly improved performance on coding benchmarks, reaching 55.8% on Terminal-Bench 4.0 (compared to 42.0% for Opus 5 and 60.9% for the specialized Mythos variant), along with strong results on agentic scientific research (52.6% on Terminal-Bench-Science 0.1), computer use (77.9% on OSWorld 2.0 strict), and business workflows (31.4% on AutomationBench). The model costs approximately 25% less than Fable 5 for typical workloads due to 75% reduced pricing on cache reads, with savings up to 45% for highly agentic tasks. Fable 5.1 is now available on all platforms including AWS, Google Cloud, and Microsoft Azure, priced at $10 per million input tokens and $50 per million output tokens, while Claude Mythos 5.1 offers more permissive safeguards for vetted cybersecurity and life sciences professionals.

MiniMax released Hailuo H3 (Hailuo 3.0) on July 31, 2026, advancing its video generation model to native 2K resolution at 24fps with clips extended to 5-15 seconds, up from previous 1080p limits. The key innovation is synchronised audio generated in a single pass alongside video, eliminating separate sound post-production. H3 introduces an omni-reference control system accepting text, image, video, and audio inputs together, and offers instruction-based editing. On the HuggingFace community leaderboard, H3 ranked second for text-to-video with an Elo rating of 1,242. MiniMax open-sourced H3's base weights (33B transformer) on August 3, 2026 via Hugging Face and ModelScope under the MiniMax H3 Community License, while keeping prompt preprocessing and 2K upscale modules available through API only.