GPT-6 Sol leaks, screenshots and release rumors

OpenAIDevs said GPT-6 Sol (xhigh) leads Fable 5.1 on AutomationBench, which tests real business workflows across apps, at 88% lower reported cost per task.

Every's team compared GPT-6 Sol with Opus 5.5: writing close to Astra, computer use matching Astra on 17 of 18 attempts, coding better than 5.6 but still behind Opus 5.5 on long autonomous builds.

The post says GPT-6 Sol and Luna are out, bringing Astra-level capabilities to faster, cheaper models, and also mentions Opus 5.5.

OpenAIDevs says GPT-6 Sol (max) nearly matches Claude Fable 5 (xhigh) on the DeepSWE v1.1 long-horizon engineering benchmark at roughly 80% lower cost per task.

OpenAIDevs says caching and inference improvements let GPT-6 Sol and Luna offer more intelligence at half the previous generation's API prices.

A Chinese long post says OpenAI released GPT-6 Sol and Luna about 90 minutes after Claude Opus 5.5, with API prices cut in half and roughly half as many factual errors as the previous generation.

An emotional post says all the earlier leaks, backend changes and routing anomalies were real, and that OpenAI cut prices sharply.

A screenshot shows GPT-6 Sol priced at roughly half of Opus 5.5 on standard input/output tokens, and the poster complains Anthropic only ships expensive models.

A meme image jokes about the two labs trading launches the same day: Anthropic released Opus 5.5 and OpenAI immediately followed with GPT-6 Sol and GPT-6 Luna.

The post links to openai.com and says Sol and Luna join the GPT-6 family, beating GPT-5.6 while costing 50% less.

The account says both new models are a decent upgrade and priced at roughly half their predecessors, with an image that appears to show pricing or benchmark results.

The official launch post positions Sol for coding and computer use and Luna as the low-cost option, with AutomationBench and OSWorld 2.0 comparisons and per-million-token prices.