Jun 29, 2026 11:20:00

OpenAI announces the 'GPT-5.6' series, surpassing Claude Mythos 5, but releasing it as a limited preview due to US government directives.

OpenAI announced the GPT-5.6 series on June 26, 2026. The top-of-the-line GPT-5.6 Sol boasts performance exceeding that of Anthropic's top-of-the-line model, Claude Mythos 5, but at the time of writing, it is not yet publicly available and is being released to a limited number of organizations.

Previewing GPT-5.6 Sol: a next-generation model | OpenAI

https://openai.com/index/previewing-gpt-5-6-sol/

GPT-5.6 Preview System Card - OpenAI Deployment Safety Hub
https://deploymentsafety.openai.com/gpt-5-6-preview

The GPT-5.6 comes in three versions: the flagship 'GPT-5.6 Sol,' the balanced model for everyday use 'GPT-5.6 Terra,' and the lower-priced 'GPT-5.6 Luna.' All three models are currently available in limited preview to a select group of trusted partners whose participation has been shared with the U.S. government, and are expected to be generally released within a few weeks.

The graph below shows the benchmark results for 'GPT-5.6 Sol,' 'Claude Mythos 5,' 'GPT-5.6 Terra,' 'Claude Fable 5,' 'GPT-5.5,' 'GPT-5.6 Luna,' 'Claude Opus 4.8,' and 'Gemini 3.1 Pro Preview.' In

Terminal-Bench 2.1 , which measures the ability to perform practical tasks on the terminal, GPT-5.6 Sol recorded a higher score than Claude Mythos 5. GPT-5.6 Terra also showed a score comparable to Claude Fable 5, and the lower-cost model GPT-5.6 Luna also outperformed Claude Opus 4.8 and Gemini 3.1 Pro Preview.

In

ExploitBench , which measures cyberattack capabilities, GPT-5.6 Sol achieved a score equivalent to Claude Mythos Preview with one-third the number of output tokens.

The GPT-5.6 series is also characterized by its low cost. The fee per 1 million tokens is as follows:

	input	output
GPT-5.6 Sol	$5 (approximately 809 yen)	$30 (approximately 4853 yen)
GPT-5.6 Terra	$2.50 (approximately 404 yen)	$15 (approximately 2426 yen)
GPT-5.6 Luna	1 dollar (approximately 162 yen)	$6 (approximately 971 yen)

The following are the results of running

GeneBench v1 , a benchmark test measuring the biological quantitative analysis capabilities of 'GPT-5.6 Sol,' 'GPT-5.6 Terra,' 'GPT-5.6 Luna,' and 'GPT-5.5.' The horizontal axis shows the API fee, and the vertical axis shows the score. It can be seen that GPT-5.6 Sol and GPT-5.6 Terra show higher scores at lower fees than GPT-5.5.

OpenAI reportedly provided the U.S. government with a briefing on the GPT-5.6 series' capabilities prior to its official release. They also explained that the limited release to select organizations before general availability was 'at the request of the U.S. government.' OpenAI commented, 'Such government access programs should not become the long-term standard. Government access programs deprive users, developers, businesses, cyber defense personnel, and global partners of the best tools.'

OpenAI also stated, 'We are taking this short-term measure to make (the GPT-5.6 series) more widely available in the coming weeks. We will work with the administration to develop a framework for executive orders and a repeatable process for future model releases,' making it clear that they aim to build an AI model release method that does not require limited preview releases.