OpenAI Unveils GPT-5.6 Sol, Boasting Unprecedented Speed and Benchmarks Amidst Government Access Controls

OpenAI's Latest Flagship: GPT-5.6 Sol Unveiled

OpenAI has officially launched a limited preview of its new flagship model, GPT-5.6 Sol, positioning it as a direct competitor to Anthropic's Claude Mythos class of models. This next-generation AI is currently accessible to a select group of partners through its API and Codex platform. The rollout comes with a new, layered naming convention: the 'x.6' denotes the generation, while 'Sol' signifies the flagship performance tier. Other planned tiers include 'Terra,' designed to match GPT-5.5's capabilities at half the cost, and 'Luna,' a budget-friendly option. Beyond these, OpenAI also outlined a 'Max' mode for deeper reasoning and an 'Ultra' mode, which is designed to delegate complex tasks to multiple sub-agents operating in parallel.

Unprecedented Speed and Performance Benchmarks

A significant highlight of the GPT-5.6 Sol announcement is its projected speed. OpenAI plans to launch GPT-5.6 Sol on Cerebras in July, promising speeds of up to 750 tokens per second (tps). This rate is considered an unprecedented velocity for a frontier intelligence model, with initial access limited to select customers as capacity expands. Industry observers suggest that such a rapid response time could dramatically enhance the utility of AI agents, particularly for demanding tasks like navigating complex codebases. Benchmarking data released by OpenAI indicates Sol's strong performance against competitors. On Terminal-Bench 2.1, a measure for agentic coding, GPT-5.6 Sol achieved a score of 88.8 percent. The even more advanced Sol Ultra variant, though not yet released, reached 91.9 percent on the same benchmark, surpassing Claude Mythos 5's 88 percent and Fable 5's 84.3 percent. Furthermore, Sol demonstrated improvements in biology, scoring 30 percent on GeneBench v1 for genomics and quantitative biology, compared to GPT-5.5's 22 percent, while consuming fewer tokens. In cybersecurity, specifically on ExploitBench, which assesses an AI's ability to find and exploit security flaws in Google's V8 JavaScript engine, Sol matched the performance of Mythos Preview, notably using approximately one-third fewer output tokens.

Government Oversight and OpenAI's Frustration

The deployment of GPT-5.6 Sol is occurring under specific directives from the U.S. government, which dictates customer-by-customer approval for access. This regulatory oversight extends to the broader AI landscape, as evidenced by the government's previous intervention to disable Anthropic's Claude Fable 5 and Mythos 5 models for all customers globally. OpenAI has openly voiced its dissatisfaction with this level of government control, stating that such processes "shouldn't become the long-term default." The company argues that these restrictions prevent essential tools from reaching "users, developers, enterprises, cyber defenders, and global partners who need them," hindering the widespread adoption and benefit of advanced AI technologies.

The Tease of Sol Ultra

Adding to the anticipation, Thibault Sottiaux, an engineering lead on OpenAI's Codex team, recently teased the unreleased GPT-5.6 Sol Ultra model. Sottiaux encouraged users to "stash your hardest prompts somewhere," indicating the advanced capabilities of this future iteration. While the Ultra variant has no official timeline, features, or rollout details announced by OpenAI, it appears to be distinct from the currently launched limited preview of GPT-5.6 Sol. The tease has generated considerable excitement, with users expressing eagerness for its potential in creative agent swarms and complex coding applications, though some also voice frustration over the prolonged hype without immediate availability.

The ongoing tension between rapid AI advancement and government oversight will likely define the accessibility and practical application of frontier models like GPT-5.6 Sol.