The AI Catchup - September 29, 2026

OpenAI DevDay 2026, the launches that matter:

• GPT-6.1 Sol: matches GPT-6 Astra on coding, says OpenAI, at a fifth of the price ($2/$10 per million tokens)
• dots: always-on ChatGPT agents that keep working between conversations
• Pro 500: $500 a month for 25x Plus usage and Astra Ultrafast, up to 8x faster in Codex
• The catch: Pro 200 drops to 10x Plus from 20x, per OpenAI's Tibo Sottiaux

Every claim below is fact-checked in the linked write-ups.

GPT-6.1 Sol: Astra-Level Coding at Sol's Price

Before: 74.1% on DeepSWE meant GPT-6 Astra at $10/$50. Now: GPT-6.1 Sol scores 75.2% at $2/$10.

• Cost per task: about $0.65 on DeepSWE, against Astra's $4.43
• Vs GPT-6 Sol: better best score on all six benchmark charts OpenAI published
• Cached input: $0.10 per million tokens, half of GPT-6 Sol's and a tenth of Astra's
• Effort: high is the sweet spot for coding; medium already scores 73.0% for $0.42
• Breaking changes: none effort is gone, and Chat Completions has no tool calling
• Where Astra still wins: computer use (+2.1 points), business workflows (+5.3), science (+11.1)

Scores and costs are OpenAI's. Every effort level and the API specs.

dots: An Agent That Keeps Working Between Conversations

Before: agents ran when you asked. Now: a dot takes a standing goal and keeps working on its own cloud computer.

• Runs on: GPT-6 Astra, with memory from the apps you connect
• Hands off code: creates cloud tasks in your existing Codex environments
• You set the autonomy: custom rules for some actions, from "act without asking" to "hand off to you"
• Who gets it: Pro (outside the EEA, Switzerland, and the UK) and Business Premium, 18+, rolling out gradually
• Cost: off the meter for the first month; OpenAI has not said what comes after
• Watch out: disconnecting an app does not delete what your dot learned from it

ChatGPT Space launched alongside it as a shared home for Pages, files, and your dot. Setup, billing, and how dots compare.

Pro 500 and Ultrafast: OpenAI Now Sells Speed

Before: Codex topped out at Fast mode. Now: Astra Ultrafast runs up to 8x Standard token speed, about 300 tokens per second.

• Speed costs allowance: Ultrafast burns included usage at 8x the Standard rate, then credits at 6x
• Token speed, not task time: runs dominated by tests and tool calls will not finish 8x sooner
• API: $60 input and $300 output per million tokens, service_tier: "ultrafast"
• Pro 200: every Pro tier now costs $20 per Plus-sized allowance; the old Pro 200 was $10
• Our call: pay $500 if you use more than half the old Pro 200 allowance or wait on Astra all day; otherwise stay

The full plan breakdown, with Claude Max compared.

Also From DevDay

• Codex CLI: start and steer tasks by voice; a new /agents view tracks several at once
• Code review: summaries, diffs, and automatic cloud first passes in the ChatGPT desktop app
• Agents API: now supports computer use, plus Codex's multi-agent tools and context compaction
• Sign in with ChatGPT: Plus and Pro allowances work in 16 partner tools, including Devin, Notion, and Vercel

Source: OpenAI's DevDay recap.

Quick Hits

• Claude Sonnet 5.5: same $2/$10 price as Sonnet 5, up to 30% lower cost per task, already in Cursor
• Claude Code evals: /claude-api build-eval and /claude-api hillclimb improve an app against held-out tests
• Cursor Rollouts: a Teams and Enterprise bot that flags deploy regressions from logs, metrics, and traces
• Claude plugin directory: paid-plan developers can submit MCP connectors and plugin bundles

Ship It This Week

1. Move coding agents from gpt-6-sol to gpt-6.1-sol at high effort, after fixing none effort and Chat Completions tool calls.
2. Rerun your Astra coding evals on GPT-6.1 Sol; keep Astra for computer use and science.
3. With dots access, give it one recurring job while usage is off the meter.
4. Pull your September Codex usage and pick Pro 200 or Pro 500 before October 29.

Account and Billing Notes

• Pro 200 grandfathering: eligible existing subscribers keep the old allowance through October 29, 2026; it does not add Ultrafast. Details.
• Claude refusal billing: Anthropic bills pre-output refusals again in three safeguard categories. Which ones.

---

The flagship is no longer the default: GPT-6.1 Sol does most of Astra's coding work for a fifth of the price, and OpenAI now charges extra for Astra's speed. If a teammate is still paying Astra prices for everyday coding, forward this.

Until next week, stay caught up.