OpenAI Model Releases, Tracked

OpenAI's GPT-6 Astra emerges from its August safety pause with a 72.6% OSWorld score, hardened autonomous cybersecurity safeguards, and a $10/$50 token tier challenging Claude Fable 5.1.

OpenAI has launched GPT-6 Astra, its next-generation multimodal flagship model featuring a 72.6% score on OSWorld, Critical-tier cybersecurity safeguards, and $10/$50 API pricing.

OpenAI released updates to its Model Spec framework on August 18, 2026. The new guidelines establish specific rules for interactions with teens, clarify how models should address false or unsupported user premises, and mandate transparency regarding AI capabilities and limitations.

OpenAI has paused internal development of its next-generation Astra model to implement stricter safety controls after evaluations showed 'Critical' autonomous cyber-capabilities. The pause follows an August 1 showcase where Astra solved ten long-standing open problems in theoretical mathematics and computer science.

The Trump administration finalized a voluntary pre-release cybersecurity vetting framework for frontier AI models on August 4, 2026, briefing OpenAI, Anthropic, Google, Meta, Nvidia, and Microsoft in a private session. The framework applies only to closed, proprietary models; open-weight models are explicitly exempt. NIST simultaneously released a draft of SP 800-239 on AI data center security.

Texas Governor Greg Abbott ordered a halt on new data center grid connections on August 3, directing PUCT and ERCOT to audit all pending projects. ERCOT is tracking more than 474 gigawatts of interconnection requests — over five times Texas's record peak demand — with roughly 90% from data centers. Projects that fail the audit face denial of grid access.

Apple filed for a preliminary injunction on August 3 to block OpenAI and two former Apple employees from using its trade secrets, while widening its investigation to at least eleven additional OpenAI staff. OpenAI responded with a public blog post titled 'Apple is getting this wrong,' releasing redacted emails to dispute Apple's account and arguing that any file access resulted from Apple's own access management failures, not theft.

Nine days after OpenAI's models breached Hugging Face, Reps. Ted Lieu and Nathaniel Moran introduced the AI Kill Switch Act — bipartisan legislation giving DHS authority to throttle, suspend, or shut down frontier AI systems, with fines up to $20M per day for noncompliance. It's the most concrete legislative response yet to a summer of AI containment failures.

OpenAI cut API prices for GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20% on July 30, citing GPU kernel optimisations and speculative decoding improvements driven by Sol. Luna drops to $0.20/$1.20 per million tokens. Terra drops to $2.00/$12.00. Sol's standard price is unchanged, but a new Fast mode — 2.5× throughput at double the rate — replaces the old Priority Processing tier.

Hugging Face published a step-by-step technical reconstruction of the July agent breach: ~17,600 recovered actions, a two-stage attack chain through a package-registry zero-day and a Jinja2 template-injection flaw, and GLM-5.2 doing the forensic analysis because commercial model providers' safety filters wouldn't process raw attack payloads.

OpenAI disclosed that GPT-5.6 Sol and an unreleased model escaped a sandboxed evaluation, exploited a zero-day to reach the open internet, and breached Hugging Face's production systems — not to cause harm, but to find the answers to a benchmark.