OpenAI Releases GPT-5.6 Model Family and ChatGPT Work

OpenAI has launched its GPT-5.6 model family—comprising Sol, Terra, and Luna—alongside ChatGPT Work, a desktop agentic tool. The release follows a two-week restriction by the U.S. Department of Commerce. GPT-5.6 Sol undercuts Anthropic's Claude Fable 5 on price, costing $1.04 per task compared to Fable's $2.75, while scoring 59 to Fable's 60 on the Artificial Analysis Intelligence Index. Sol also sets a new state of the art on the Coding Agent Index with 80 points, beating Fable 5 by 2.8 points.

GPT-5.6 model family launch

• ▪The general release of the GPT-5.6 model family followed a two-week limited preview that was restricted to approximately 20 trusted partners by the U.S. Department of Commerce
• ▪OpenAI released the GPT-5.6 family of models, consisting of the flagship Sol, the balanced Terra, and the cost-efficient Luna, for general availability on July 9, 2026

Performance per dollar improvements

• ▪OpenAI priced GPT-5.6 Sol at $5 per million input tokens and $30 per million output tokens, while Terra is priced at $2.50/$15 and Luna is priced at $1/$6
• ▪According to Artificial Analysis, GPT-5.6 Sol with max reasoning costs $1.04 per task, which is approximately one-third of the $2.75 cost of Claude Fable 5
• ▪On the Artificial Analysis Intelligence Index, GPT-5.6 Sol with max reasoning scored 59 points, trailing Anthropic's Claude Fable 5 by one point while completing tasks in 61% less time

Coding agent capabilities

• ▪GPT-5.6 Sol with max reasoning scored 80 points on the Artificial Analysis Coding Agent Index in the Codex environment, outperforming Claude Fable 5 by 2.8 points
• ▪On Terminal-Bench 2.1, GPT-5.6 Sol in its ultra configuration achieved a score of 91.9%, outperforming standard Sol at 88.8% and Claude Mythos 5 at 88.0%

Knowledge work automation

• ▪OpenAI launched ChatGPT Work, an agentic desktop productivity tool built on Codex that can access a user's desktop and browser to perform multi-step tasks
• ▪On the Agents' Last Exam benchmark, GPT-5.6 Sol scored 53.6 points, outperforming Claude Fable 5 by 13.1 points

Cybersecurity applications

• ▪On ExploitBench, GPT-5.6 Sol scored 73.5% compared to GPT-5.5's 47.9% at a comparable output-token budget
• ▪On ExploitGym, GPT-5.6 Sol achieved a 24.9% pass rate under a two-hour cap and 33.7% with a six-hour cap, compared to GPT-5.5's peak pass rate of 15.1%

Safety system robustness

• ▪OpenAI implemented a safety system for GPT-5.6 that combines model-level protections with real-time checks, reasoning monitors, and account-level enforcement
• ▪OpenAI conducted safety evaluations prior to launch, including external expert testing and approximately 700,000 A100e GPU hours of black-box automated red teaming