OpenAI moved the GPT-5.6 family to general availability on Thursday, following a limited preview that began June 26 restricted to vetted partners.

The family is three named tiers — Sol as flagship, Terra as everyday, Luna as cheapest — rolling out across ChatGPT, Codex and the API at $5/$30, $2.50/$15 and $1/$6 per million input/output tokens. It shipped alongside ChatGPT Work, an agentic workspace that pulls context from Slack, Notion, Microsoft 365, Google Drive and local files and produces documents, spreadsheets, presentations and hosted sites. OpenAI claims state of the art on the Artificial Analysis Coding Agent Index at 80, 2.8 points above Claude Fable 5, and 53.6 on Agents' Last Exam. Reuters reported the earlier delay came at the request of the Trump administration on national-security grounds, with the Commerce Department's Center for AI Standards and Innovation clearing the broad rollout before launch.

Meta Superintelligence Labs released Muse Spark 1.1, a multimodal reasoning model built for agentic work with a 1-million-token context window, native multi-agent orchestration — it can act as a primary agent delegating to parallel subagents, or as a subagent itself — and computer use across desktop applications.

Alongside it Meta opened a public preview of the Meta Model API, OpenAI-compatible, US developers only, with $20 in free credits, at $1.25 per million input tokens and $4.25 per million output tokens. Meta says Muse-family models are expected to replace the Llama models powering assistants in WhatsApp, Instagram, Facebook and its smart glasses. Early partners named in the launch post include Replit, Cline and Box.

OpenAI released GPT-Live-1 and GPT-Live-1 mini, a third generation of ChatGPT voice models built on a full-duplex architecture that processes input while generating output, making interaction decisions many times per second and handling interruptions and live translation.

GPT-Live-1 becomes the default for paid ChatGPT tiers and GPT-Live-1 mini for free users, replacing Advanced Voice Mode globally across iOS, Android and ChatGPT.com. When a request needs search or deeper reasoning, GPT-Live delegates to a frontier model — GPT-5.5 at launch — asynchronously while the conversation continues. API access is planned, and OpenAI published a system card with voice-native safety evaluations the same day.

Elon Musk said Wednesday that xAI was making Grok 4.5 available to the public; Reuters and trade coverage date the public release to July 8, with EU access deferred.

It is xAI's first model built from the ground up for coding and agentic tasks rather than general chat, trained alongside Cursor, with a 500,000-token context window, configurable reasoning effort, roughly 80 tokens per second serving speed and pricing at $2/$6 per million input/output tokens. It is the default model in Grok Build and available across every Cursor plan and the xAI API console. Musk framed it as an Opus-class model, but faster, more token-efficient and lower cost.

Anthropic published a case study of the Government of Alberta's Ministry of Technology and Innovation, which used Claude Code with Claude Opus and Sonnet across the systems of all 27 provincial ministries.

Around 50 agents ran autonomously and in parallel through a two-stage routine — a rules engine to flag known vulnerability patterns, then review that cites the exact file and line for every finding — covering roughly 1,280 applications and 3,400 code repositories in about 20 hours. The ministry estimates the equivalent manual review would have taken about 6.5 years. Claude Code also generated, tested and built fixes, and wrote tests where none existed. Alberta published 21 technical papers on the method and plans to consolidate 185 legacy applications into 16 modern ones.