Breaking

OpenAI Opens Limited Preview of GPT-5.6 Series, With Flagship 'Sol' Targeting Coding and Agentic Work

July 17, 2026 at 17:00 EDT

Foundation Models
AI Agents
Software Dev & Coding

OpenAI on July 9, 2026 launched a limited preview of its GPT-5.6 model series, led by the flagship GPT-5.6 Sol, which the company positions as its strongest model for complex coding, science, cybersecurity and long-horizon agentic tasks. Broader access is planned for the coming weeks.

July 9, 2026 · OpenAI

GPT-5.6 Sol Enters Preview as OpenAI's Strongest Coding & Agent Model

A three-tier family — Sol, Terra and Luna — leads on complex coding, science and long-horizon agentic tasks, pairing capability gains with a hardened safety stack and lower cost per task.

91.9%

Sol Ultra on Terminal-Bench 2.1 (agentic coding)

Token context window across all three tiers

~⅓

Cost per task vs top Claude models at comparable intelligence

TERMINAL-BENCH 2.1 — AGENTIC CODING

Higher score = stronger command-line & agentic coding

91.9%

Sol Ultra

88.8%

Sol (base)

88.0%

Claude Mythos 5

83.4%

GPT-5.5

GPT-5.6 SOL

Flagship

Coding, science & long-horizon agentic work · $5 / $30 per M tokens

GPT-5.6 TERRA

Balanced

GPT-5.5-level quality at roughly half the cost

GPT-5.6 LUNA

Fast / Cheap

Lowest latency and price in the series

EARLY PRACTITIONER RESPONSE

Analyzed a ~15,000-line Python codebase, found 47 improvements and implemented them in a single short session — praised for persistent, long-sequence execution and creative consistency.

STILL OPEN

Narrow limited-preview access means broad, independent testing is pending. Key questions: do the coding & agentic leads hold up, and do cost-efficiency claims drive lasting adoption?

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.