OpenAI Releases GPT-5.6 Model Family, Citing Performance Gains and Cost Efficiency

OpenAI announced on Thursday the general availability of its GPT-5.6 family of artificial intelligence models, led by the new flagship model, Sol. The launch, which follows a limited preview, positions Sol as a high-performance option that the company claims outperforms key competitors on capability and cost-efficiency.

Competitive Benchmarks

OpenAI released several benchmark results comparing Sol to rival models, including Anthropic's Claude Fable 5. The company stated that GPT-5.6 Sol demonstrates significant advantages in performance, speed, and cost, particularly in complex, agentic tasks.

Key performance claims from OpenAI include:

• Professional Workflows: On the Agents’ Last Exam evaluation, Sol achieved a score of 53.6, surpassing Claude Fable 5 by 13.1 points.
• General Intelligence: On the Artificial Analysis Intelligence Index, Sol with max reasoning came within one point of Fable 5 but completed tasks in 61% less time at roughly half the estimated cost.
• Coding Capabilities: For coding, Sol scored 80 on the Artificial Analysis Coding Agent Index, 2.8 points higher than Fable 5, while using less than half the output tokens and costing about one-third less, according to the company.

Itamar Friedman, CEO of the code-review platform Qodo, stated that GPT-5.6 was the strongest model his company had evaluated on its agentic tests, noting it beat OpenAI's own GPT-5.5 while using approximately three times fewer tokens and delivering about two times lower median latency.

Model Tiers and Pricing

Ad

The GPT-5.6 family includes three distinct models to serve different use cases and budgets. Alongside the flagship Sol model, OpenAI introduced Terra, a balanced model for general work, and Luna, its most cost-efficient offering.

The company also introduced a new "ultra" capability setting that coordinates four agents in parallel to deliver stronger results on demanding tasks more quickly, albeit with higher token consumption. API pricing per 1 million tokens has been set as follows:

• Sol: $5 for input and $30 for output.
• Terra: $2.50 for input and $15 for output.
• Luna: $1 for input and $6 for output.

Safety and Availability

OpenAI reported that the models underwent extensive safety testing, including approximately 700,000 A100e GPU hours of automated red teaming. While the company noted that GPT-5.6 models are more capable in sensitive fields like biology and cybersecurity than previous versions, it said they do not cross its defined "Critical" capability threshold.

The GPT-5.6 models became available on Thursday across ChatGPT, Codex, and the OpenAI API. The company said the rollout is beginning globally and will proceed gradually toward full availability over the next 24 hours.