OpenAI launches GPT-6 Sol and Luna models in Codex and ChatGPT Work.

// 4h agoMODEL RELEASE

OpenAI launches GPT-6 Sol and Luna models in Codex and ChatGPT Work.

OpenAI has announced the availability of two new tier models, GPT-6 Sol and GPT-6 Luna, within OpenAI Codex and ChatGPT Work. Sol acts as a robust workhorse for complex agentic workflows and coding tasks, while Luna provides a faster, cost-effective alternative for focused, high-volume subagent work, both expanding on the capabilities introduced by GPT-6 Astra.

// ANALYSIS

The introduction of Sol and Luna showcases OpenAI's strategy to segment its flagship models based on task complexity and cost-efficiency.

–Sol is positioned to tackle heavy lifting in development environments, optimizing for complex multi-step planning and tool validation.
–Luna offers a highly efficient alternative for routine tasks, reducing overhead for high-volume API and coding workflows.
–The dual release, paired with token price reductions, suggests an aggressive push to dominate both enterprise development (Work tier) and broader autonomous agent ecosystems.

// TAGS

openaigpt-6codexaillm

DISCOVERED

4h ago

2026-09-22

PUBLISHED

4h ago

2026-09-22

RELEVANCE

10/ 10

AUTHOR

mark_k

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

UPDATE23m ago

Grok Bot desktop app ships 53 performance improvements

The desktop application for Grok Bot has shipped 53 performance optimizations over the past few days to dramatically reduce latency across core workflows. Key improvements include cutting network reconnection times after a flaky connection from 60 seconds to 0.7 seconds, reducing laptop wake recovery from 23 seconds down to 1 second, and bringing the transition time when switching back to the Computer view down to 868 milliseconds.

INFRA1h ago

Boat ships small 5GHz agent sandboxes

Boat has rolled out its small sandbox tier featuring 2 vCPUs clocked at 5GHz, 4GB RAM, and 12GB disk for $0.000005 per second ($0.018/hour). The persistent Ubuntu environments target AI agent workloads with zero-disk-cost snapshots, preinstalled devtools, and concurrency scaling up to 2,000 instances.

INFRA1h ago

Astra finds 14-year-old libuv memory leak

After investigating intermittent ChatGPT desktop crashes on macOS, developer Peter Steinberger prompted AI agent Astra to inspect system crash logs and trace native stack traces down to a 14-year-old memory leak in libuv's Darwin filesystem watcher. Pull request #5283 resolves the bug by releasing uncollected Core Foundation objects and C path buffers across exit paths in `uv__fsevents_reschedule`, halting quadratic memory growth during watch churn.

OpenAI launches GPT-6 Sol and Luna models in Codex and ChatGPT Work.