GPT-5.6 Launches Under US Oversight; MCP Enterprise Auth Goes Stable

<p><strong>GPT-5.6 Sol, Terra, and Luna</strong> open to public API access today — the first frontier models to clear the new US Commerce Department security review — bringing explicit cache breakpoints and a three-tier pricing structure ($1–$5/M input). Meanwhile, <strong>MCP Enterprise-Managed Authorization</strong> reaches stable, delivering zero-touch...

Highlights

  • GPT-5.6 Sol, Terra, and Luna open to public API access today after the first full US Commerce Department frontier AI security review — establishing a new 1–2 week vetting lag baseline for future frontier model releases (OpenAI)
  • MCP Enterprise-Managed Authorization (EMA) reaches stable with Okta as first IdP and day-one adoption from Anthropic, Microsoft VS Code, Atlassian, Figma, Linear, and Supabase — zero-touch SSO for every connected MCP server (MCP Blog)
  • The MCP 2026-07-28 release candidate is now published: stateless core, MCP Apps (sandboxed HTML tool UIs), Tasks API, and a formal deprecation policy — 19 days for SDK maintainers to validate (MCP Blog)
  • Gemini 3.1 Flash-Lite preview (gemini-3.1-flash-lite-preview) shuts down today — any call still using the preview alias will hard-fail; migrate to gemini-3.1-flash-lite immediately (Google)

Key Signals

ranked scan
  1. 1

    GPT-5.6 Sol Goes Public Under the New US Frontier AI Oversight Framework

    July 9, 2026 — After two weeks of restricted access to ~20 Commerce Department-approved partners, Sol ($5/$30 per 1M tokens), Terra ($2.50/$15), and Luna ($1/$6) open broadly via API today. Sol ships explicit cache breakpoints and a 30-minute minimum cache life — a long-requested feature for large RAG pipelines — and is the first frontier model to formally clear the Trump administration's new Center for AI Standards and Innovation vetting process. (OpenAI, The Next Web)

    models
  2. 2

    MCP Enterprise-Managed Authorization Is Stable

    June–July 2026 — EMA solves the OAuth onboarding friction that stalled enterprise MCP rollouts: an IT admin authorizes servers once via the identity provider, and users inherit access automatically on first login — no per-app OAuth, nothing to configure per user. Okta is the launch IdP; Anthropic surfaces it across Claude, Claude Code, and Cowork; VS Code ships it in-IDE. Asana, Atlassian, Canva, Figma, Linear, and Supabase are live on day one, with Slack support actively in progress. (MCP Blog, InfoQ)

    auth
  3. 3

    MCP 2026-07-28 RC: What Actually Changes

    July 2026 — The published release candidate formalizes stateless protocol core (no sticky sessions, plain round-robin load balancing, Mcp-Method header routing), MCP Apps (SEP-1865, server-hosted sandboxed HTML UIs hosts can security-review ahead of time), a Tasks API for long-running work, and a deprecation policy that replaces ad-hoc breaking changes. Tier 1 SDKs are expected to ship support within the ten-week window; July 28 is the hard compatibility marker for production deployments. (MCP Blog)

    spec

Why It Matters / What To Watch

operator actions

  1. US Government AI Oversight Is Now an Ops Variable

    • GPT-5.6 is the first model to complete the new Commerce Department review — future frontier releases from any lab will carry a similar pre-release vetting lag; plan preview-to-GA upgrade windows with that buffer in mind. (The Next Web)
    • The 19-day Fable 5/Mythos 5 shutdown (restored July 1) showed that export-control events can sever enterprise access without warning; teams without model-provider fallback routing should treat that as a live gap, not a tail risk. (CNBC)

  2. Three MCP Enterprise Deadlines Are Converging This Month

    • EMA stable + RC published + July 28 spec freeze means the next 19 days are the optimal window to validate MCP integrations, enroll enterprise IdPs, and pin SDK versions before the spec locks. (MCP Blog, MCP RC)
    • Builders on Google's Flash tier: gemini-3.1-flash-lite-preview dies today — swap to gemini-3.1-flash-lite, which carries the same $0.25/$1.50 pricing with a 2.5× TTFT improvement over 2.5 Flash. (Google)