gg2

OpenAI releases GPT 5.6 Sol vision model

Archived — this story has rotated out of today’s deck. It is kept here in full.

The gist

OpenAI released GPT-5.6 Sol, its best vision model, scoring 46.2 mAP@50 in detection. The model also powers new cybersecurity and ultrafast modes, signaling a broader AI push.

Background

OpenAI announced the GPT-5.6 lineup last week, introducing Sol, Terra, and Luna models with a focus on computer use and visual understanding. Independent benchmarks from Roboflow show Sol is the best vision model OpenAI has released, with significant improvements in object detection and counting. The model also serves as the base for specialized variants like GPT-5.6-Cyber for cybersecurity and an Ultrafast mode for faster processing.

How it unfolded

  1. Jul 16, 2026Roboflow published a blog post benchmarking GPT-5.6 Sol, Terra, and Luna, highlighting Sol's superior vision capabilities.
  2. Aug 10, 2026OpenAI announced GPT-5.6-Cyber, a cybersecurity-focused model built on GPT-5.6 Sol, and expanded its Daybreak program with Blue and Red tiers.
  3. Aug 13, 2026OpenAI introduced Ultrafast mode for GPT-5.6 Sol, offering up to 14x faster processing with 750 output tokens per second.

Who’s saying what

Analysts
Roboflow's benchmark shows Sol is the best vision model OpenAI has released, with major gains in detection and counting.
Caution
OpenAI confirmed Sol becomes less stable on images around 2,000 by 2,000 pixels or larger, especially at lower reasoning effort.

Still unverified

Roboflow's benchmark is independent but not yet publicly released; Zhipu's claims about GLM-5.3 are based on its own testing.

Sources

See today’s stories in the app gg2 — free on the App Store