OpenAI Releases GPT-6.1 Sol With Lower Pricing

OpenAI released GPT-6.1 Sol this week, an updated version of its mid-tier model within the GPT-6 family. The company claims GPT-6.1 Sol achieves performance comparable to its higher-tier GPT-6 Astra model in tasks such as agentic coding, computer use, and professional work, but at one-fifth the token price. This new model is available immediately through the OpenAI API under the identifier gpt-6.1-sol, and is integrated into ChatGPT Work and Codex. The pricing structure for the GPT-6 family now includes three tiers: GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna. GPT-6 Astra is priced at $10 per million tokens for input and $50 for output, with a $1 per million token rate for cached input. GPT-6.1 Sol is priced significantly lower at $2 for input, $10 for output, and $0.10 for cached input per million tokens. The most affordable tier, GPT-6 Luna, costs $0.10 for input, $0.50 for output, and $0.01 for cached input per million tokens. The reduction in cached input costs is particularly beneficial for AI agents, which frequently re-process the same system prompts, tool schemas, and conversational history. For GPT-6.1 Sol, cached reads now represent only 5% of the uncached input rate, a substantial decrease from the 10% rate associated with the previous GPT-6 Sol model. OpenAI has provided benchmark results, sourced from their launch post, comparing GPT-6.1 Sol against other models. In coding tasks on the DeepSWE v1.1 benchmark, GPT-6.1 Sol reportedly matches GPT-6 Astra's performance at approximately one-fifth the cost and surpasses the best score of GPT-6 Sol by 6.4 percentage points, while requiring less reasoning effort. For professional work, tested on the GDP.pdf benchmark which evaluates responses to complex professional documents, GPT-6.1 Sol is stated to outperform Claude Opus 5.5 with fallbacks at less than half the cost per task. On the AutomationBench 1.0.6, Sol scores 2.2 points higher than Opus 5.5 at medium effort, at roughly one-third the cost. In computer use scenarios, evaluated on the OSWorld 2.0 offline dataset, Sol shows a 7-point improvement over GPT-6 Sol at maximum effort, and approaches Astra's performance, achieving within 2.1 points at approximately one-seventh the cost per task. For scientific applications, the Terminal-Bench Science 0.1 benchmark shows GPT-6.1 Sol more than doubling the score of GPT-6 Sol at maximum effort. The average cost per task for Sol in this domain is $5.47, compared to $23.21 for Opus 5.5 and $23.80 for Astra. Despite these advancements, OpenAI recommends GPT-6 Astra for the most demanding research tasks, where it leads with a 68.1% score. The company also noted factuality improvements at low.