In late June 2026, OpenAI opened a limited preview of GPT-5.6 Sol, a new flagship model built for cybersecurity defense, alongside its related security platform Daybreak. Early access is restricted to a small group of vetted organizations, including U.S. government-approved partners.
Frontier AI · Private & Limited Releases
Frontier Labs Pivot to Tiered, Restricted-Access Models
xAI quietly opens Grok 4.5 —a 1.5-trillion-parameter model—as a private beta inside SpaceX and Tesla, while OpenAI debuts the GPT-5.6 family and a security toolkit to limited partners only. The era of one all-purpose model is giving way to use-case-specific, gated rollouts.
1.5T
Parameters in Grok 4.5's V9 base — about 3× its predecessor
91.91%
GPT-5.6 Sol on TerminalBench 2.1 (ultra)
3
GPT-5.6 tiers: Sol, Terra & Luna for different workloads
Daily
Updates to the Grok Build code/agent harness via RL
Parameter Scale: Grok 4.5 vs. Predecessor
Roughly 3× the parameter count — each block ≈ 0.5T parameters
~1.5T
Grok 4.5
xAI · Grok 4.5
1.5T-param V9 base + Cursor supplemental training
Early internal evals: on par with / exceeding Claude Opus
Tested in live SpaceX & Tesla engineering settings
Pledge: a freshly trained model every month through 2026
Access: Private beta, internal only
OpenAI · GPT-5.6 + Daybreak
Sol — complex coding & security researchTerra — high-volume business tasksLuna — fast everyday tasksDaybreak — vuln. verification, patch gen, risk triage
Access: U.S. gov-approved partners
What draws attention
Massive 1.5T scale and Cursor integration signal direction; OpenAI's security specialization promises real-world utility in vulnerability fixing.
Why caution remains
No independent third-party benchmarks for Grok 4.5 yet; tight access restrictions are seen as a barrier to broader adoption.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…