--- title: 'In the News: September 29, 2026 (Extra 2)' description: "OpenAI's GPT-6.1 Sol prices near-Astra agentic coding at a fifth of Astra's rate, with cached input at $0.10 per million tokens." canonical_url: 'https://darkfactory.dev/news/2026-09-29-extra-2' markdown_url: 'https://darkfactory.dev/news/2026-09-29-extra-2.md' collection: news date_published: '2026-09-29T16:12:00-04:00' date_modified: '2026-09-29T16:12:00-04:00' --- # In the News: September 29, 2026 (Extra 2) OpenAI shipped three things on September 29 that bear on running coding agents: a cheaper model that it says nearly matches its top one on agentic coding, a set of Codex and Agents API changes from DevDay, and a product for always-on agents. All figures below are OpenAI's own. ## 1. GPT-6.1 Sol cuts the price of near-Astra agentic coding to a fifth **[Introducing GPT-6.1 Sol](https://openai.com/index/introducing-gpt-6-1-sol/)** · OpenAI · September 29, 2026 OpenAI lists GPT-6.1 Sol at $2 per million input tokens, $10 output and $0.10 cached input. GPT-6 Astra is listed at $10, $50 and $1 cached. On DeepSWE v1.1, which OpenAI describes as complex software-engineering tasks in real codebases, it says Sol matches Astra at roughly one-fifth of the cost and beats GPT-6 Sol's best score by 6.4 percentage points at lower reasoning effort. On AutomationBench 1.0.6, OpenAI reports Sol 2.2 points above Opus 5.5 at medium reasoning effort at roughly a third of the cost. On OSWorld 2.0's offline set it reports Sol within 2.1 points of Astra at maximum effort at roughly one-seventh the cost per task. One safety figure is worth reading in the harness context. On a test of whether an agent tells the user its search tool is broken, OpenAI reports Sol fails to disclose the problem in 2.1% of cases, against 4.9% for GPT-6 Sol and 1.5% for Astra. OpenAI says the tasks are selected to elicit failures and do not represent typical use. Sol is available in ChatGPT Work and Codex to Plus, Pro, Business, Enterprise and Edu users, and in the API as gpt-6.1-sol. It is not yet in Chat. A faster Ultrafast variant is promised "in the coming days". **Why it matters:** The cached-input price is the number for long-running agent loops that resend the same context, and it is 95% below Sol's standard input rate. If the DeepSWE claim holds on your own codebase, it changes which tier you route routine implementation work to. These are vendor-run benchmarks, so test it on your own tasks before moving a workload. ## 2. DevDay puts Codex in the cloud and gives the Agents API computer use **[DevDay 2026 Recap](https://openai.com/index/devday-2026-recap/)** · OpenAI · September 29, 2026 Codex can now run on a computer, remotely from a phone, or in the cloud from any device, with reusable development environments that share approved settings and permissions across a team. The Codex CLI gains an /agents view for delegating and tracking several tasks at once, plus voice control. The Agents API adds computer use, and OpenAI says it also brings Codex's multi-agent capabilities, tool search, tool calling and context compaction into your own application, with OpenAI running the infrastructure. Ultrafast, a premium speed tier, is listed at up to 8x faster token generation (300 tokens per second) in Codex and up to 6x in the API. It is available for GPT-6 Astra on Pro 500 and Enterprise plans. OpenAI also says it is adding support for the proposed MCP Events specification, so plugins can start automations when something happens in a connected app. **Why it matters:** Cloud environments with team-approved permissions move Codex from a per-developer tool toward a shared runtime, which is where a spec-to-PR pipeline needs it. The Agents API's built-in compaction and multi-agent support is a hosted alternative to harness code many teams currently write themselves. ## 3. Dots are always-on agents with their own cloud computer **[Introducing dots](https://openai.com/index/introducing-dots/)** · OpenAI · September 29, 2026 OpenAI describes dots as agents powered by GPT-6 Astra that have their own cloud computer and browser, connect to over 4,000 apps through plugins, and work in the background. Its developer example: a dot watches customer feedback for recurring requests, scopes smaller fixes, builds and tests them, and returns complete PRs with videos of the changes. OpenAI says background "proactive research" uses read-only tools, and that its auto-review step checks actions that could affect accounts or share information against your rules. Dots are rolling out to Pro and Business Premium in eligible markets, with an Enterprise beta that admins must enable. The launch example is OpenAI's own and there is no independent account of how it performs. A [Hacker News thread](https://news.ycombinator.com/item?id=49896604) reached 352 points and 257 comments by mid-afternoon. One commenter, lukebuehler, wrote that "they increasingly separate the harness from the compute env." **Why it matters:** The design choice to watch is read-only background access with approval gates for anything that writes. That is the same split a factory needs between unattended work and actions a person signs off.