At its global DevDay conference, artificial intelligence research leader OpenAI officially unveiled its latest software architecture, debuting persistent "Dots" AI agents alongside the high-performance GPT-6.1 Sol foundation model. Moving beyond traditional prompt-and-response chatbot interfaces, the newly introduced Dots framework enables persistent software assistants to operate continuously across background tasks, execute complex computer operations, and integrate directly with multi-platform enterprise tools such as Slack, Microsoft Teams, and cloud computing environments. Operating on an expansive 1-million-token context window, the underlying GPT-6.1 Sol model allows developers and enterprises to analyze massive code repositories and extensive technical documentation in a single operational pass. OpenAI executives highlighted that the new model costs approximately one-fifth of previous flagship architectures, drastically lowering operational expenditures for startups and global enterprises deploying automated workflows. Technology analysts and enterprise software architects observe that releasing affordable, background-capable AI agents accelerates the enterprise shift toward autonomous digital workforces, fundamentally changing how organizations automate software development, complex data processing, and day-to-day administrative operations.

Paradigm Shift: Continuous Persistent Execution Replaces Single-Prompt Chat

In a major architecture update to its platform, OpenAI introduced Dots—always-on, persistent AI agents that continuously handle ongoing goals and background responsibilities. Departing from traditional turn-based chat loops, a Dot operates independently within its own isolated cloud environment and browser, maintaining long-term context across apps like Slack, Microsoft Teams, and ChatGPT Work.

Alongside the Dots agent framework, OpenAI released GPT-6.1 Sol, an upgraded intermediate frontier model designed specifically to run long-horizon agentic workloads, repository-level software engineering, and multi-step business automation at scale.

Overview: Key Specifications, Benchmarks & Capabilities of OpenAI Updates

Feature / Model ParameterOperational Architecture & SpecificationsPrimary Use Cases & Impact
OpenAI 'Dots' AgentsAlways-on background execution; cloud browser & computer accessProject tracking, continuous monitoring & automated execution
GPT-6.1 Sol Context Window1,050,000 Input Tokens • 128,000 Max Output TokensLarge codebase analysis, long-form document synthesis
Software Engineering BenchmarkMatches GPT-6 Astra on DeepSWE v1.1Complex repository refactoring & multi-file coding
Factuality & Error RateFactual error rate reduced to 7.7% (vs 11.4% in GPT-6 Sol)Accurate document research & enterprise automation
API Pricing Structure$2.00 / 1M Input • $0.10 / 1M Cached Input • $10.00 / 1M Output~80% cost reduction compared to GPT-6 Astra

GPT-6.1 Sol: Near-Astra Intelligence at One-Fifth the Cost

The GPT-6.1 Sol model fills the gap between the ultra-high capability GPT-6 Astra and the lightweight GPT-6 Luna. By offering near-Astra level reasoning across software development, document parsing (GDP.pdf), and computer-use automation (OSWorld 2.0) at 20% of Astra's token price, Sol provides developers and enterprises a viable economic pathway to run continuous agent workloads.

OpenAI Agentic Architecture Workflow: ------------------------------------- Continuous User Goal ──> 'Dots' Persistent Agent Monitor ──> Cloud Sandbox / Browser Execution ──> GPT-6.1 Sol Low-Cost Reasoning Loop

In agentic coding evaluations on DeepSWE v1.1, GPT-6.1 Sol matched the benchmark scores of GPT-6 Astra while reducing per-task compute costs significantly.

Enterprise Deployment, Workspace Integration, and Safety Protocols

OpenAI is deploying Dots across Pro, Business, and Enterprise tiers within ChatGPT Work and dedicated Codex workspaces. To enforce safety standards, Dots operate under defined authority boundaries—separating read-only research or local drafting from high-consequence actions like external publishing or financial transactions, which require explicit human-in-the-loop sign-off.

API access for gpt-6.1-sol is available immediately for developers, with an upcoming Ultrafast variant tailored for high-throughput code synthesis.