GPT-6.1 Sol

Overview

GPT-6.1 Sol is a model released by OpenAI on September 29, 2026, at DevDay 2026, succeeding GPT-6 Sol, which was only online for 7 days. It is officially positioned to approach GPT-6 Astra in agentic coding, computer use, and professional work tasks, while standard input and output token prices are about one-fifth of Astra's. The model ID is gpt-6.1-sol, it is available through the API, and it is open to Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex; regular Chat is not yet available. Also released at the same event was the Astra Ultrafast acceleration tier, with up to 8x generation speed in Codex and up to 6x in the API, at about 300 tokens per second, priced at 6x the standard rate.

Key Features

• Agentic Coding and Professional Task Positioning: It is officially positioned to approach GPT-6 Astra in agentic coding, computer use, and professional work tasks, while standard input and output token prices are about one-fifth of Astra's.
• Ultra-Long Context: The context length is 1.05 million tokens, suitable for handling large-scale codebases and long-document analysis tasks.
• Cached Input Discount: The discount for cached input relative to standard input is 95%, so the cost of reusing context is far lower than resending it.
• Lower Error Rate at Low Reasoning Tier: At the low reasoning tier, the proportion of responses containing errors or hallucinations dropped from 11.4% in the previous generation to 7.7%, leaving only a 1.9 percentage point gap from Astra's overall error rate.
• Multiple Access Points: It is available through the API and is open to Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex; regular Chat is not yet available.
• Acceleration Tier Released at the Same Event: The Astra Ultrafast acceleration tier released at the same event offers up to 8x generation speed in Codex and up to 6x in the API, at about 300 tokens per second, priced at 6x the standard rate.

Use Cases

• Software engineering tasks: capability on DeepSWE v1.1 is close to Astra, with a cost of about $1.5 per task, versus about $7.7 for Astra.
• Professional knowledge work such as PDF document analysis: overall performance exceeds Claude Opus 5.5, while task cost is less than half of the latter.
• Agentic coding and computer use: officially positioned to cover agentic coding, computer use, and professional work tasks.
• Long-context reuse scenarios: cached input discount is 95%, suitable for conversations and document processing that require repeated context reuse.
• High-throughput generation scenarios: can be paired with the Astra Ultrafast acceleration tier, with up to 8x generation speed in Codex and up to 6x in the API.

Pros

• Approaches GPT-6 Astra in agentic coding, computer use, and professional work tasks.
• Standard input and output token prices are about one-fifth of Astra's.
• The discount for cached input relative to standard input is 95%, so the cost of reusing context is far lower than resending it.
• Context length is 1.05 million tokens.
• On DeepSWE v1.1, the cost is about $1.5 per task, versus about $7.7 for Astra.
• On professional knowledge work evaluations such as PDF document analysis, overall performance exceeds Claude Opus 5.5, while task cost is less than half of the latter.

Pricing

API pricing is $2 per million tokens for input, $0.10 for cached input, and $10 for output; Astra is correspondingly $10 for input, $50 for output, and $1 for cached input, with a 95% discount for cached input relative to standard input. The Astra Ultrafast acceleration tier released at the same event is priced at 6x the standard rate. Pricing information not provided is subject to the official website.

Summary

GPT-6.1 Sol is a model released by OpenAI on September 29, 2026, at DevDay 2026, succeeding GPT-6 Sol, which was only online for 7 days. It is officially positioned to approach GPT-6 Astra in agentic coding, computer use, and professional work tasks, while standard input and output token prices are about one-fifth of Astra's. The model ID is gpt-6.1-sol, it is available through the API, and it is open to Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex; regular Chat is not yet available.

Version History

• GPT-6.1 Sol default context length increased to 500k. (2026-10-12): OpenAI has increased the default context length of GPT-6.1 Sol in Codex and ChatGPT Work from 272k to 500k, an increase of about 80%, while the maximum limit remains 1M. The company says this value has been thoroughly tested and can reduce how often context compression is triggered, and users can still override it manually. On the same day, it announced that calling GPT-6.1 Sol from subscription quotas will no longer incur long-context fees.
• GPT-6.1 Sol's default context length in Codex and Work has been increased to 500k. (2026-10-11): Tibo announced that the default context length of GPT-6.1 Sol in Codex and Work has been increased from 272k to 500k, an increase of about 80%. This can reduce the occurrence of compaction and help tasks that need to retain a large amount of context and details over the long term; users can still override the setting themselves, and the model's maximum context length is 1M.
• Sol (Max) enters Agent Arena at No. 5 (2026-10-03): Arena reported GPT-6.1 Sol (Max) at No. 5 on Agent Arena with a +11.23% net improvement and a median cost of $0.56 per task, saying it reshaped the Pareto frontier of the leaderboard
• GPT-6.1 Sol (Max) enters Agent Arena at 5th place, approaching the top models at a lower cost. (2026-10-02): Arena announced that OpenAI's GPT-6.1 Sol (Max) ranks 5th (+11.23%) in the Agent Arena, with a median task cost of $0.56, and has reshaped the Pareto frontier.
• Coding Agent Index cost figures (2026-10-01): Artificial Analysis data puts GPT-6.1 Sol at about 0.72 dollars per task, roughly 30 percent below GPT-6 Sol at 1.05 dollars, which was already half of GPT-5.6 Sol at 1.99 dollars. GPT-6.1 Sol also ranks near the top of the Artificial Analysis Coding Agent Index
• Code Arena ranking and price-performance (2026-09-30): GPT-6.1 Sol (Max) took third place on the Code Arena: WebDev leaderboard with 1759 points; third-party data including Artificial Analysis showed Sol approaching Astra on coding and computer use at roughly one fifth of the price
• OpenAI releases GPT-6.1 Sol, replacing GPT-6 Sol after only seven days (2026-09-29): It approaches GPT-6 Astra on agentic coding, computer use and professional tasks at roughly one fifth of Astra standard input and output token prices: 2 dollars per million input, 0.10 dollars cached and 10 dollars output, with a 1.05 million token context window.