OpenAI releases GPT-6.2 Astra and Sol with lower API prices and benchmark gains

OpenAI has introduced two new language models, GPT-6.2 Astra and Sol, offering strong benchmark performance and reduced API costs.

2 min read

OpenAI releases GPT-6.2 Astra and Sol with lower API prices and benchmark gains

OpenAI has released two new language models named GPT-6.2 Astra and GPT-6.2 Sol. Astra is designed for demanding professional work in fields like science, law, and finance, while Sol is built for quick, everyday tasks. Both models are available now for Plus, Pro, Enterprise, and Edu users.

Key facts

Astra input price
$8 per million tokens
Astra cached input price
$0.40 per million tokens
Astra output price
$40 per million tokens
Sol input price
$2 per million tokens
Sol cached input price
$0.10 per million tokens
Sol output price
$10 per million tokens
GPT-6.2 model comparison
Primary use
GPT-6.2 Astra: Demanding professional work
GPT-6.2 Sol: Everyday tasks
Input price per million tokens
GPT-6.2 Astra: $8
GPT-6.2 Sol: $2
Output price per million tokens
GPT-6.2 Astra: $40
GPT-6.2 Sol: $10
Bar chart and scatter plot illustrating the Intelligence Index rankings and cost per task across language models.
Artificial Analysis Intelligence Index showing GPT-6.2 Astra ranking second with strong cost-performance efficiency., Source ↗

Performance and Benchmarks

Independent evaluations show mixed results across different domains. In the Artificial Analysis Intelligence Index, GPT-6.2 Astra ranks second behind Anthropic’s Claude Fable 5.2, but it leads the cost-performance ratio. On scientific and multi-step professional tasks, Astra outperforms its predecessor while using fewer output tokens. However, coding evaluations show flat results, and legal tests reveal that both models trail behind specialized competitors.

Line graph comparing model scores and costs on the Agents' Last Exam v1.1 benchmark.
Agents' Last Exam v1.1 benchmark results showing scores relative to cost for GPT-6.2 Astra and competing models., Source ↗

Pricing and Availability

Developers can access both models through the OpenAI API. API pricing for GPT-6.2 Astra is set at eight dollars per million input tokens, forty cents for cached inputs, and forty dollars per million output tokens, representing a twenty percent reduction compared to its predecessor. GPT-6.2 Sol maintains the same pricing as the previous version, charging two dollars per million input tokens, ten cents for cached inputs, and ten dollars per million output tokens.

Safety and Governance

Both models reach the Critical threshold for cybersecurity in OpenAI's Preparedness Framework, demonstrating an increased ability to find and exploit vulnerabilities. In response, OpenAI is introducing stricter safeguards for cybersecurity requests. Additionally, Astra is the first externally released model to reach OpenAI's High threshold for AI self-improvement, which the company monitors using internal safety systems.

Source: Original announcement

Worth a read?

Report

Comments