This is one of Roadmap's ready-made topic templates — a starting point for tracking something over time without writing a tracking brief from scratch.
Roadmap is an AI-assisted document editor: markdown with a live preview, a discussion panel where you can ask the AI about a specific section and apply its suggested edit straight to the text (you choose exactly what goes into each request's context), version history with restore, and export in multiple formats. Its "AI Update" button, on any document, searches the web for what has changed since you last checked and updates the document — with dates and sources — only when it finds something genuinely new.
To use this template: click "Open in Editor" above to copy it into your own documents, then press "AI Update" whenever you want to check for news. Browse the sidebar for more topics and categories.
AI Model Releases
Brief
Tracking new releases of large language models (OpenAI, Anthropic, Google, Meta, Mistral, DeepSeek, and others). Focus on: release dates, claimed benchmarks, API pricing, context limits. Not needed: general speculation about the future of AI, analyst opinions without concrete numbers.
Description
The frontier AI market has become a genuine multi-way race rather than a single clear leader, with pricing compressing sharply and open-source models steadily closing the gap — even as export-control policy increasingly complicates which models are actually available where.
Model Landscape in Mid-2026
July 2026 has been the most turbulent month yet for frontier AI availability. On June 12, a US export-control directive forced Anthropic to suspend Claude Fable 5 and Mythos 5 globally; the directive was lifted on July 1, restoring global access to Claude Fable 5. On June 26, OpenAI launched its next generation, GPT-5.6, as a three-model family (Sol, Terra, and Luna), initially limited to a preview with about twenty partner organizations. On June 30, Anthropic launched Claude Sonnet 5.
Benchmarks and Pricing
Claude Sonnet 5 scores 63.2% on SWE-bench Pro and is priced at $2/$10 per million input/output tokens through August 31, 2026. OpenAI's Sol variant scores 91.9% on TerminalBench 2.1, while the cheaper Luna variant scores 82.5% at $1 per million input tokens. OpenAI says its mid-tier Terra model matches GPT-5.5 at roughly half the cost.
Open-Source Models
Among openly downloadable models, GLM-5.2 currently scores highest on general intelligence benchmarks, while DeepSeek V4 and Kimi remain favorites for reasoning and coding tasks respectively.