Breaking

OpenAI Made 'Many Changes' in Talks With Government Before GPT-5.6 Public Launch

July 9, 2026 at 11:06 EDT

Foundation Models
Policy & Regulation

OpenAI Chief Executive Officer Sam Altman said the company made "many changes" during discussions with the Trump administration before releasing its newest GPT-5.6 models to the general public. Responding to the administration's cybersecurity concerns, OpenAI had shifted its originally planned broad availability to a staggered rollout.

June 26, 2026 · OpenAI

GPT-5.6 Debuts on a Government-Vetted Leash

Sam Altman says OpenAI made "many changes" after talks with the Trump administration, swapping a broad launch for a staggered rollout — first to ~20 approved partners, then public within weeks. Altman calls it "not the preferred long-term model."

~20

approved partners with initial "government-vetted access"

models — Sol, Terra & Luna

stake OpenAI is reportedly discussing offering the U.S. government

TerminalBench 2.1 — coding & agent benchmark

Higher is better. Sol leads the frontier field.

91.9%

Sol
Ultra Mode

88.8%

Sol
standard

88.0%

Claude
Mythos 5

70.7%

Gemini
3.1 Pro

The three models · price per 1M tokens

Sol — flagship $5 in / $30 out

Complex tasks, coding & security research · Ultra Mode ~750 tokens/sec on Cerebras

Terra — balanced $2.50 in / $15 out

GPT-5.5-class performance at half the price · everyday work

Luna — fastest $1 in / $6 out

Cheapest · summarization & routine work

Supporters say

Terra delivers strong performance at low cost
Sol's subagents shine on complex tasks
The "best tools" shouldn't be kept from users

Critics warn

Benchmark reliability questioned; flagged cheating in METR evals
Government-vetted access could become an enduring default
Deepening ties between AI firms and the administration

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.