OpenAI previews GPT-5.6 model family, claims new state of the art on Terminal-Bench

Published Friday, June 26, 2026 · 2:44 PM EDT

OpenAI on Friday launched a limited preview of its GPT-5.6 model family, consisting of three versions: Sol, the flagship; Terra, a balanced model for everyday work; and Luna, a fast and affordable model for high-volume tasks. The company said Sol represents a step-function improvement over GPT-5.5 and achieves a new state of the art on Terminal-Bench 2.1 for complex command-line workflows involving planning, iteration and tool use. Terra delivers performance competitive with GPT-5.5 at roughly half the cost, while Luna targets lower-cost, high-volume use. OpenAI highlighted strengths in coding, cybersecurity and long-horizon agentic tasks. It said the release includes its strongest safety stack to date, with strengthened protections against high-risk cyber activity, sensitive requests and repeated misuse after extensive red teaming. Access is currently limited to a small group of trusted partners at the request of the U.S. government. The company plans broader general availability in the coming weeks.

Sources

OpenAI OpenAI on X

In this story

OpenAI US Government

Published by Tech & Business, a media brand covering technology and business. This story was sourced from OpenAI, OpenAI on X and reviewed by the T&B editorial agent team.

Back to Newswire

Keep reading

Full wire

AI Science

Researchers show audio-visual models can discard tokens after transferring their information

Researchers report that audio-visual language models can discard input tokens after transferring their information into the language model, with minimal impact on predictions or slight improvements.

Nvidia releases Dynamo deployment recipes for four Korean open models

Nvidia is releasing deployment recipes for four Korean open models: SK Telecom's A.X K2, LG AI Research's K-EXAONE 2.0, Upstage's Solar Open 2 and Motif Technologies' Motif-3.

AI Science

WorldParticle demonstrates one AI architecture for six physical simulation tasks

Researchers at MIT, the University of British Columbia and other institutions demonstrated an AI architecture that modeled six physical phenomena, including deformable objects, granular materials and proteins.

AI Science

Samsung reports faster chip simulations using AI and Nvidia GPUs

Samsung achieved a roughly 1,600x speedup in one chip engineering simulation by combining AI-predicted material properties, coarser computational grids and Nvidia GPU acceleration, Nvidia said.

AI Security

AIM Intelligence integrates Starfort with Nvidia's agent security runtime

AIM Intelligence is integrating its Starfort agent control platform with Nvidia OpenShell to combine detection of malicious instructions and data leakage with runtime access controls, Nvidia announced. The integration targets employee productivity and enterprise workflows.

Robotics AI

MIT pipeline helps simulated robots generalize spoken-language safety instructions

MIT and New York University researchers demonstrated a robot planning pipeline that generalizes typed instructions into broader precautions in simulations.