OpenAI’s release of GPT‑6.1 Sol brings a mid‑tier model that matches the high‑end GPT‑6 Astra on many benchmarks while cutting costs dramatically. For developers building AI agents, the biggest pain point is the recurring expense of resending the same system prompt, tool schemas, and context on every step. GPT‑6.1 Sol lowers the cached‑read price from $0.20 to $0.10 per million tokens, halving the cost of that repetitive work. In a typical agent loop—30 steps per task, 60 K token prefix reused 90 % of the time, 1.5 K output tokens per step—this translates to roughly a 50 % reduction in total spend compared with the previous GPT‑6 Sol model.
Performance wise, vendor‑reported benchmarks show GPT‑6.1 Sol nearing Astra levels: it ties Astra on DeepSWE v1.1 coding at about one‑fifth the price, beats Opus 5.5 on GDP.pdf professional‑work tests at less than half the cost, gains 7 points over GPT‑6 Sol on OSWorld 2.0 computer‑use tasks while staying within 2.1 points of Astra at roughly one‑seventh the cost, and more than doubles GPT‑6 Sol’s score on Terminal‑Bench Science 0.1. Factuality improves too, with error rates dropping from 11.4 % to 7.7 % at low effort—a 32 % reduction.
Deploying the model is straightforward: it is available as a hosted model via the OpenAI API under the identifier gpt-6.1-sol, and it appears in ChatGPT Work and Codex. The API supports a 1,050,000‑token context window, 128,000 max output tokens, text and image input, and a range of reasoning.effort settings (low, medium‑default, high, xhigh, max). Tool calling should use the Responses API; plain chat completions work without it. Note that prompts exceeding 272 K input tokens incur 2× input/cache rates and 1.5× output rates, batch and Flex modes are 50 % cheaper, and fast mode doubles cost (unavailable with EU residency). Fine‑tuning is not supported.
For teams focused on cost‑effective agents, coding assistants, or professional‑work automation, GPT‑6.1 Sol offers a practical path to near‑top‑tier performance without the premium price. Start by prototyping with the default medium reasoning effort, monitor cached‑read savings, and scale up to high or max effort only when the task demands it.
#AI #Product #Development #Automation #LLM #CostEfficiency