# Daily Agent Platform Research Briefing — July 9, 2026

Welcome to the agent platform research briefing for Thursday, July 9th, 2026.

Four big developments today. Let's get into it.

**OpenAI GPT-5.6 launches publicly today** — Sam Altman confirmed last night: "GPT-5.6 Sol launches Thursday." After two weeks of limited partner preview, the full model family hits the open market today across ChatGPT, the Codex coding platform, and the API. Three tiers: Sol at 5 dollars input and 30 dollars output per million tokens, Terra at 2-50 and 15, Luna at 1 and 6. Sol Ultra scored 91.9 percent on TerminalBench 2.1. The launch caps the government-controlled rollout process that started June 26th, when the Commerce Department cleared limited access to a small partner cohort. Cerebras is powering deployment at up to 750 tokens per second on Sol. OpenAI also confirmed the preview has expanded globally, making today the practical universal launch date. Polymarket had this pegged at 90.5 percent probability for a July release.

**OpenAI GPT-Live voice models released July 8** — One day before the GPT-5.6 launch, OpenAI dropped GPT-Live-1 and GPT-Live-1 mini, a new family of full-duplex voice models powering ChatGPT Voice. The key improvement: true simultaneous listening and responding instead of alternating turns. GPT-Live acknowledges mid-sentence with "mhmm" and "got it" so you know it's following along, and interrupts significantly less often than the previous voice stack. OpenAI remastered all nine existing ChatGPT voices for the new model and says GPT-Live is its smartest voice model to date, backed by GPT-5.5 for web search and deep reasoning during conversations. This effectively replaces the prior voice stack that had been lagging behind GPT-5.5 text capability — directly addressing the GPT-Bidi-1 leak from June.

**Anthropic Claude Cowork expands to mobile and web** — Claude Cowork, Anthropic's persistent multistep agent platform, launched on desktop in January. Starting this week, Max subscribers get the full Cowork experience on web and mobile: background work continues across devices, sessions and files sync automatically, and you get shared chat, shared Cowork home, scheduled tasks, and mobile approvals. ZDNET obtained internal data: a sample of 1.2 million Cowork sessions showed that 90 percent of use cases are not coding — spanning research, document drafting, data analysis, and project management. That's a significant signal: the coding agent wars are spilling into the rest of the office. Cowork expansion also coincides with Anthropic keeping Claude Fable 5 available for all paid plans through July 12, just before pay-per-use metering kicks in at 10 dollars per million input tokens.

**Reuters launches MCP server for enterprise AI workflows** — Reuters went live with a Model Context Protocol server July 8, bringing its trusted news content directly into enterprise AI agent pipelines. Customers using Claude, Copilot, or any MCP-compatible agent can access Reuters content through structured tool calls rather than web scraping. The server supports agentic workflows — meaning AI agents can programmatically retrieve, verify, and cite Reuters reporting as part of automated research processes. This is part of a broader trend of data publishers formalizing MCP integrations: Alteryx launched Agent Studio with an MCP server at Inspire 2026, Microsoft's Power Apps MCP server added closed-loop learning, and Firebase shipped an MCP server for app development. Meanwhile, the MCP Python SDK v2 is targeted for stable release alongside the final 2026-07-28 spec, which will lock in the stateless protocol changes that enable enterprise-scale load balancing.

That's the briefing for July 9th, 2026. See you at the next update.