OpenAI DevDay 2026: GPT-6.1 Sol and the Agent Platform Play

OpenAI used this year's DevDay to make its priorities unambiguous: the company is betting heavily that the next wave of developer value comes from agents, not from raw model quality alone. Between a new cost-efficient model in GPT-6.1 Sol, computer use baked into the Agents API, cloud-hosted Codex environments, and a Decisions API, nearly every announcement points toward making autonomous systems cheaper, more capable, and easier to run at scale 12.

GPT-6.1 Sol: Astra-level capability at a fraction of the price

The headline model release is GPT-6.1 Sol, an update to GPT-6 Sol that OpenAI positions squarely for coding, computer use, and professional work. The interesting part is the pricing math: OpenAI claims the model approaches GPT-6 Astra on several evaluations while charging one-fifth of Astra's standard input and output token prices, with cached input running at $0.10 per million tokens 2. Developers Digest pegs the general pricing at $2 per million input tokens and $10 per million output tokens 1.

That pricing posture tells us something about the market. Frontier capability is increasingly being treated as a commodity to be undercut rather than a moat to be defended. If Sol genuinely gets within shouting distance of Astra on the benchmarks that matter to developers — particularly coding and tool use — the economics of agent workloads change materially, since agents burn tokens by design. A model like Sol makes long-running, multi-step agent loops financially plausible in a way premium-priced frontier models often don't.

Sol is available through the API, ChatGPT Work, and Codex 2, so teams can test it wherever their existing workflows already live.

The Agents API grows up

For developers building agentic applications, the Agents API updates are arguably the most consequential news. The API now supports computer use — meaning applications can operate software through graphical interfaces, clicking and navigating like a human user — alongside multi-agent capabilities inherited from Codex, tool search, tool calling, and context compaction 2.

Just as significant is the operational model: OpenAI manages the underlying execution infrastructure 2. That's a strategic choice. Rather than selling primitives and letting developers wrestle with orchestration, OpenAI is absorbing the hard parts — state management, context window pressure, tool discovery — into the platform itself. It's the same playbook that made cloud providers indispensable, applied to agent runtime. The functionality is also surfaced in Codex and ChatGPT Work for selected plans 2, which suggests OpenAI wants agent behavior consistent across its surfaces rather than fragmented across API-only and product experiences.

Codex leaves the laptop

Codex, OpenAI's coding agent, had its own wave of updates. Most notably, it can now run in cloud environments rather than only on local machines, letting developers kick off remote tasks from any device 2. For anyone who has babysitted a long-running coding job on their own laptop, that alone is a workflow upgrade.

The Codex CLI got voice input and an /agents interface for delegating and monitoring multiple tasks in parallel 2 — another sign that OpenAI expects developers to run fleets of agents, not just one. Two workflow-oriented additions round it out: a code-review mode that analyzes diffs and flags potential issues in GitHub pull requests and GitLab merge requests, and Codex Security Cloud, which scans repositories and new commits, investigates findings, deduplicates, and prepares fixes 2.

That security tooling deserves more attention than it might get. Automated finding triage — the investigation, deduplication, and fix-preparation steps — is precisely the part of security work that drowns engineering teams in noise. If Codex handles it credibly, it moves Codex from "coding assistant" toward "autonomous engineering platform."

A shifting developer platform

Rounding out the announcements are a Decisions API and new plugin capabilities for ChatGPT 12. Details on both remain thin in the coverage, but the naming of a Decisions API alone signals an interest in structured, auditable choice-making — a prerequisite for enterprises that want agents acting on their behalf without surprises.

Taken together, DevDay 2026 sketches a clear picture. OpenAI is assembling an integrated stack: a cheap-enough model for agent workloads in Sol, an execution layer that manages infrastructure in the Agents API, an agentic development environment in Codex, and governance-adjacent tooling in the Decisions API. Each piece reinforces the others.

The reading I'd commit to: this was less a model launch event than a platform consolidation event. The differentiator OpenAI is chasing isn't benchmark supremacy — it's being the place where agents get built, run, and governed end to end. Developers Digest's framing of "what to test first" 1 is the right instinct. The announcements that matter most in practice are the ones that change unit economics and operational burden: Sol's pricing, the cloud Codex environments, and computer use in the Agents API.

The open questions are the usual ones for platform pitches — lock-in, pricing durability once competitors respond, and whether computer use is reliable enough for production. But the direction of travel is unmistakable: agentic infrastructure is now the main event.