GPT-6.1 Sol sets OpenAI record amid heavy load delays

Sam Altman acknowledged response slowdowns for the agent-focused model and promised improvements following backend infrastructure repairs.

On October 1, 2026, Sam Altman posted on X that GPT-6.1 Sol had become the fastest-growing model in OpenAI history. In the same message, the chief executive acknowledged that the system struggled under heavy traffic (“a bit slow under load”) and stated that backend adjustments had just been deployed to improve response times (“Should be much better now!”).

The surge began right after the model debuted at DevDay on September 29. Drawn by lower API rates and specialized coding capabilities, engineering teams moved batch workloads to the new model immediately. That sudden wave of requests placed heavy pressure on OpenAI’s server infrastructure.

Quick recap

Sam Altman stated GPT-6.1 Sol set an OpenAI adoption record, though absolute traffic numbers remain undisclosed.
Pricing at $2.00 per million input tokens cuts inference costs to roughly one-fifth of GPT-6 Astra.
Technical threads on Hacker News and LinkedIn documented notable latency during the first 48 hours.
OpenAI began rolling out an Ultrafast mode designed to generate tokens up to eight times faster.

What GPT-6.1 Sol is and why demand surged

Introduced at DevDay 2026, GPT-6.1 Sol was designed for agentic software engineering, external tool use, and automated enterprise workflows. Unlike typical chatbots built for single questions and answers, software agents chain dozens of consecutive calls to inspect code repositories, execute terminal commands, and fix bugs autonomously.

To drive developer adoption, OpenAI set low rates across the API: $2.00 per million input tokens, $0.10 per million cached tokens (a 95 percent discount), and $10.00 per million output tokens. Tokens are fragments of words used to measure language processing. That pricing provides reasoning close to the frontier GPT-6 Astra model at approximately 20 percent of the cost.

Access opened immediately under the identifier gpt-6.1-sol across the API, Codex, and ChatGPT Work for paid subscribers. The combination of reduced cost and wide availability prompted organizations to route active pipelines to the new model without a phased transition.

Heavy load slowdowns and community reports

The migration wave quickly strained available inference compute, which is the raw processing required to generate every model output. In complex agentic tasks where a system relies on a sequence of dependent steps, API delays stall the entire process and complicate interactive testing.

During the first 48 hours, discussions on Hacker News and posts on LinkedIn, including from engineer Tejas Kumar, pointed to degraded response times. Developers reported requests taking far longer than expected to complete. Altman’s public statement confirmed that the slowdown stemmed from backend cluster saturation under heavy traffic.

Benchmark claims and the Ultrafast rollout

In benchmarks published by OpenAI, GPT-6.1 Sol matched GPT-6 Astra on the DeepSWE v1.1 engineering test and scored above Claude Opus 5.5 on GDP.pdf and AutomationBench. These figures represent vendor-reported numbers without independent third-party verification, measured on evaluations selected to highlight software automation.

To mitigate delays during interactive workflows, OpenAI began rolling out a mode called Ultrafast. The company claims it generates tokens up to eight times faster for debugging and agent navigation in Codex and ChatGPT Work. For systems making dozens of sequential API calls, lower latency is essential for practical everyday use.

What OpenAI has yet to clarify

While Altman said performance should improve, OpenAI did not publish specific metrics behind its growth claim. The company did not share total API requests per second, daily token volumes, or active customer figures. OpenAI also declined to detail what technical changes were applied to its inference clusters. For teams running production agents, the real test will be whether latency stays stable during peak enterprise business hours.