OpenAI Ships GPT-5.6 Family Under Government Access Limits, Sets July/October Deprecation Deadlines

OpenAI's GPT-5.6 launch arrives under the same government-gating regime that suspended Anthropic's Fable 5, restricting the frontier tier to about 20 approved government partners for early access and vetting before any wider rollout. The three tiers are priced Sol at $5/$30 per million input/output tokens, Terra at $2.50/$15, and Luna at $1/$6, spanning heavy-reasoning to cost-sensitive workloads. OpenAI President Greg Brockman touted GPT-5.6 Sol as "a big step forward" on GeneBench-Pro, a computational-biology eval whose problems take human experts 20-40 hours.

The more immediate developer story is the deprecation wave: OpenAI set two hard deadlines, July 23 and October 23, to retire gpt-4, gpt-3.5-turbo, o1 and more than 25 legacy model IDs, alongside an August 26 Assistants API shutdown. Critics say endpoints will stop responding entirely with no soft-deprecation grace period or migration tooling.

Competitively, the government gating hands Google an opening — Gemini 3.5 Pro shipped ungated because its cyber benchmark fell below the restriction threshold — while OpenAI and Anthropic both sit behind approval walls. The pricing also undercuts the narrative that frontier costs only rise; Luna at $1/$6 is aggressive.

Community reaction to the deprecation was harsh: developers called it "a huge mistake," complaining about too little notice, no migration tooling, and undocumented behavior changes, reinforcing a "deprecation tax" narrative that "OpenAI optimises for its own product roadmap, not API consumer stability." Watch how many enterprises miss the July 23 cutoff.