Simon Willison: Previewing GPT-5.6 Sol

Main thesis

GPT-5.6 Sol matters not only for its capabilities but for its distribution model: a 'limited preview' with US-government vetting of users is a novel pattern that could shape how frontier models ship over the next several quarters.

Context

The post responds to OpenAI's June 24 Sol/Terra/Luna announcement. Willison frames it against the parallel Anthropic Mythos 5 release to US 'trusted partners' — and argues both firms are converging on a regulated-distribution path that adds legal risk for enterprise customers too.

Why it matters

For developers and CTOs: if frontier models require vetting, it's time to plan open-weight fallbacks. For policy folks: this is the first real case where a tech firm and a government share a gate-keeping role.

Details / arguments

• GPT-5.6 Sol in limited preview with government vetting of users
• Convergence with the Anthropic Mythos 5 trusted-partners model
• Pricing and full model card not yet public
• Willison recommends watching the implementation details of the access regime