OpenAI announced the release of two new large language model variants, GPT‑6.2 Astra and GPT‑6.2 Sol, on Thursday. Both models are positioned as upgrades over the previous GPT‑6 series, with Astra aimed at complex professional and scientific workloads and Sol optimized for everyday technical tasks. The rollout includes lower API pricing and expanded safety measures, according to the company’s statements.

SOL / USD
Source: CoinGecko. Prices are indicative.

Performance gains across professional benchmarks

According to CoinGape, GPT‑6.2 Astra outperformed its predecessor by 5.2 percentage points on the Agents’ Last Exam, a multistep professional‑task benchmark, and beat Anthropic’s Claude Fable 5.2 by 5.5 points. The model also posted a nine‑point advantage on the Terminal‑Bench‑Science 0.1 test and set a new high on the Vals Science Index, reflecting stronger capabilities in biological data analysis and workflow research. GPT‑6.2 Sol, while slightly behind Astra on some metrics, matched it on Cognition’s FrontierCode 2.0 and exceeded Astra on the LifeVal evaluation, which gauges real‑world daily task performance.

Reduced API costs and pricing structure

OpenAI lowered the API token price for GPT‑6.2 Astra by 20% relative to GPT‑6 Astra. The new rates are $8 per million input tokens, $0.40 per million cache tokens, and $40 per million output tokens. GPT‑6.2 Sol retains the $2 per million input and $10 per million output pricing of its predecessor, making it roughly one‑fifth as expensive as Astra and about one‑seventh the cost per task compared with GPT‑6 Astra. The pricing changes accompany OpenAI’s recent partnership with Google Cloud to deliver its models to enterprise customers.

Stricter safeguards and availability

Both models are now accessible through ChatGPT Work, Codex, and the OpenAI API for Plus, Pro, Enterprise and Edu tiers. OpenAI introduced tighter cybersecurity safeguards, classifying the models as meeting its Critical threshold for cybersecurity and High threshold for biological and chemical capabilities. High‑risk requests are restricted, though defensive research may be permitted via the Daybreak program. Monitoring systems have been upgraded to detect unauthorized actions during model operation.

What to watch next

Analysts will monitor how the pricing adjustments affect developer adoption, especially as competitors such as SpaceXAI’s Grok 4.7 and Anthropic’s upcoming releases vie for market share. Further updates on the Daybreak program and any additional safety tier adjustments are also expected as OpenAI balances capability growth with regulatory scrutiny.