OpenAI Launches GPT-6.1 Sol and Dot Agents: Performance Nears Astra, Costs Plunge

TradingKey - OpenAI announced the launch of GPT-6.1 Sol, an upgraded version of GPT-6 Sol, with a focus on enhancing agentic coding, computer use, and professional work capabilities. OpenAI stated that GPT-6.1 Sol approaches the performance of GPT-6 Astra across multiple benchmarks, while its standard input and output token prices are approximately one-fifth those of Astra. In addition, the agent Dot was also unveiled.

The standard API pricing for the model is $2 per million input tokens, $0.10 per million cached input tokens, and $10 per million output tokens. Among them, the cached input price is 95% lower than standard input, mainly targeting agentic workloads that require repeated context reuse.

GPT-6.1 Sol Targets Balance Between "High Performance and Low Cost"

OpenAI stated that GPT-6 Astra remains its frontier model with the strongest overall capabilities, suitable for the most complex scientific research and professional tasks. GPT-6.1 Sol, on the other hand, aims to provide a solution better suited for large-scale deployment by balancing performance and cost.

In the DeepSWE v1.1 code evaluation, GPT-6.1 Sol achieved performance close to GPT-6 Astra at approximately one-fifth of the cost. Under lower reasoning intensity and cost, its score was also 6.4 percentage points higher than the top performance of GPT-6 Sol.

In the GDP.pdf professional document evaluation, GPT-6.1 Sol outperformed Opus 5.5 with fallback mechanisms across different tested reasoning settings, with a per-task cost less than half that of the latter. Delivering performance close to GPT-6 Astra, its per-task cost was approximately one-fifth that of Astra.

Multiple Benchmarks Show Performance Improvement

In computer-use tasks, the score gap between GPT-6.1 Sol and GPT-6 Astra narrowed to within 2.1 percentage points, but its per-task cost is about one-seventh that of Astra.

In scientific research evaluations, the average per-task cost for GPT-6.1 Sol at maximum reasoning intensity was $5.47, lower than Opus 5.5's $23.21 and GPT-6 Astra's $23.80. GPT-6 Astra still ranked first among tested models with a score of 68.1%, and OpenAI recommends using it for the most challenging scientific research tasks.

OpenAI also disclosed that GPT-6.1 Sol recorded a factual error rate of 4.1% in high-difficulty factual accuracy tests, 0.4 percentage points lower than GPT-6 Sol, while its per-task cost was roughly 83% lower than Astra's. However, the test used filtered high-difficulty conversations and does not represent typical user scenarios.

OpenAI Launches Dot: An Always-On AI Agent

Debuting alongside Sol is the AI agent Dot. According to the announcement, Dot is powered by GPT-6 Astra, features an independent cloud computer, learns from feedback over long-term use, works around the clock toward user goals, and can connect to over 4,000 applications through a plugin ecosystem. Users can reach Dot at any time via ChatGPT, SMS, Slack, and Teams, or engage in voice calls with it, and Dot synchronizes context across all channels.

Dot features independent identity and permission management, uses a cloud computer and its own browser, and can write and test code when needed. Users can open Dot's computer at any time to inspect its work or authorize it to connect to their own laptops. Dot can autonomously advance multiple projects simultaneously without switching between different conversations or requiring step-by-step guidance. OpenAI cited an example where an early tester's Dot discovered that the user had forgotten to invoice a publisher, prepared the invoice, and sent it upon receiving approval.

Dot is being gradually rolled out to Pro, Business Premium, and Enterprise plan users in eligible markets. Meanwhile, OpenAI is opening a preview of dedicated Dots to a small number of enterprises, which hold an independent identity and clear responsibilities within the organization. Pro users can join a waitlist with limited spots to apply for sending and receiving messages via iMessage and RCS.