Sam Altman just upended OpenAI’s own release roadmap. On September 29, 2026, at the company's DevDay conference in San Francisco, OpenAI unveiled GPT-6.1 Sol and "Dots," according to The Hindu. The timing is aggressive. GPT-6.1 Sol arrives a mere week after the launch of GPT-6 Sol and Luna, and only 26 days after the flagship GPT-6 Astra made its debut.
This is not a minor incremental patch. It is a highly optimized mid-tier reasoning model designed to bring near-Astra intelligence down to a fraction of the cost. For software engineers and enterprise leaders scaling AI developer tools, the economic math of agentic coding has fundamentally shifted overnight.
The Economics of Mass-Scale Agentic Coding
High-reasoning models have historically broken developer budgets. GPT-6.1 Sol attacks this barrier directly. OpenAI has priced the model's standard API access at $2.00 per million input tokens and $10.00 per million output tokens. According to DataCamp, this represents exactly one-fifth of the pricing of OpenAI's flagship GPT-6 Astra, which sits at $10.00 input and $50.00 output.
The cost reductions run deeper for repetitive workflows. Cached input pricing has dropped to $0.10 per million tokens. This is a 95% discount compared to standard input, and a 50% reduction compared to the cached input pricing of the original GPT-6 Sol. For engineering teams running continuous integration pipelines or massive codebase refactoring loops, these margins make complex agentic workflows commercially viable.
Benchmarking Near-Astra Performance
Price cuts mean little if performance degrades. However, GPT-6.1 Sol matches or exceeds flagship-level capabilities across several key developer and reasoning benchmarks:
- DeepSWE v1.1 (Software Engineering): GPT-6.1 Sol scored 75% on this demanding benchmark, matching the performance of GPT-6 Astra at one-fifth of the cost, according to OpenAI's technical documentation. This is a notable jump from the standard GPT-6 Sol's score of 68.8%.
- GDP.pdf (Professional Documents): The model scored higher than Anthropic's Claude Opus 5.5 with fallbacks, completing tasks at less than half the cost.
- Terminal-Bench Science 0.1: At maximum reasoning effort, GPT-6.1 Sol solved tasks at a cost of $5.47 per task, compared to Astra's $23.80 per task.
- Factual Accuracy: At low reasoning effort, the model reduced factual errors by approximately 32% compared to GPT-6 Sol, with the error rate dropping from 11.4% to 7.7% as reported by The New Stack.
These metrics suggest that OpenAI is actively optimizing the compute-to-performance ratio, squeezing flagship-tier reasoning out of a significantly smaller, faster model architecture.
Architecture, Safety, and the "Dots" Integration
GPT-6.1 Sol features a 1.05-million-token context window and supports up to 128,000 tokens of maximum output. This massive output capacity is critical for agentic coding, allowing the model to generate entire multi-file pull requests in a single run.
This level of autonomy brings obvious security risks. Under OpenAI's Preparedness Framework, GPT-6.1 Sol is classified as "Critical" in cybersecurity and "High" for biological and chemical capabilities. To mitigate these risks, the model utilizes the exact same safeguards stack as the flagship GPT-6 Astra. These strict protocols are a direct response to growing industry anxieties around autonomous software agents, especially following previous containment failures on government portals.
The model is available immediately to Plus, Pro, Business, Enterprise, and Edu users within ChatGPT Work and Codex, and via the OpenAI API under the identifier gpt-6.1-sol.
Alongside the model, OpenAI introduced two major platform updates:
- "Dots" Always-On Agents: Running on secure cloud computers, these agents are powered by GPT-6 Astra. They connect to over 4,000 applications and integrate directly with Slack and Microsoft Teams to perform background tasks. You can read our deep dive on how Dots manages persistent workflows.
- "Ultrafast" Generation: Available for Codex and the API, this new tier offers up to 8x faster token generation, peaking at 300 tokens per second.
By pairing the high-speed Ultrafast engine with the low-cost GPT-6.1 Sol, OpenAI is positioning its ecosystem as the default infrastructure for the next generation of autonomous software development.
