OpenAI's new mid-priced model GPT-6 Sol ranks 7th on aggregated benchmark score

GPT-6 Sol, OpenAI's new mid-priced model, ranked seventh out of 25 models on The Latent Score, three ranks higher than its predecessor.

On Tuesday, OpenAI launched GPT-6 Sol and Luna as the cheaper distribution of the generation following Astra's launch earlier this month. The company said the two new models were trained with similar methods as Astra and inherit its advancements in professional work, factuality, coding, computer use, and alignment.

OpenAI said GPT-6 Sol can "take on difficult work tasks" while offering higher usage limits and lower cost, while Luna is the most "efficient" model for high-volume tasks. OpenAI priced Sol at $2 per one million input tokens and $10 per one million output tokens, and $0.10 input and $0.50 output charges for Luna. This is a 50% reduction from GPT-5.6 promotional rates.

The Latent Score

Upon release, GPT-6 Sol ranked seventh of 25 models on The Latent Score, which is The Latent's AI model scoring board that summarizes demonstrated capability across published reasoning settings and seven equally weighted domains. Overall, GPT-6 Sol scores 117.5, a 5.6-point jump from GPT-5.6 Sol, which sits 10th at 111.9.

GPT-6 Sol's strongest features lie in "knowledge and reasoning" as well as "agentic tool and computer use," in which the model placed sixth out of all models. It ranked seventh in "multimodal and vision," one spot behind its 5.6 model, and eighth in "coding and software engineering."

Scores are not yet available for three other domains - professional and real-world work, cybersecurity, and preference and communication. Early scores can move as more reviewed evidence arrives.

Meanwhile, GPT-6 Luna ranked 18, a 13-point rise from its previous 5.6 model.

GPT-6 Astra, OpenAI's flagship model of the generation, is currently the second-best AI model based on The Latent Score, only behind Anthropic's Opus 5.5 by five points. Astra exceeds Opus 5.5 in only one area: knowledge and reasoning.