Anthropic and OpenAI shipped new models on the same afternoon: Claude Opus 5.5 and GPT-6 Sol and Luna, both cheaper than the models they replace. These are the 8 AI and developer tools stories that got the most views on X on 22 September 2026, with the posts that carried them.

13.3M views across 5 posts

Anthropic released Claude Opus 5.5, the first model in its 5.5 family. It performs at the level of Claude Fable 5.1 on most tasks and costs 40% less to run than Opus 5. Artificial Analysis put it at the top of its Intelligence Index with a score of 58, and pricing drops to $4/$20 per 1M input/output tokens, with cache reads down from $0.50 to $0.20.

6M views across 6 posts

OpenAI released GPT-6 Sol and GPT-6 Luna, smaller models built on GPT-6 Astra, with API prices 50% below GPT-5.6. Sol drops to $2/$10 per 1M tokens and Luna to $0.10/$0.50. Artificial Analysis found cost per task roughly halved, with Sol gaining 2 points on its Coding Agent Index and Luna losing 2. Both are rolling out in ChatGPT Work and Codex for paid plans.

1.9M views across 1 post

DigitalOcean launched Managed Agents in public preview: a runtime for Claude Code, Codex or your own LangGraph agent that pauses when idle. Tools sit behind one governed endpoint, you can pick from 75+ open and proprietary models, and it bills on the same DigitalOcean account.

734K views across 1 post

Gagan Biyani launched The Horowitz Andreessen Academy in San Francisco, a school for top high school graduates built around learning to build by shipping their own projects. It raised $42M led by a16z, has 10 founding partners including Anthropic, OpenAI, Google, NVIDIA and Stripe, and opens with a 1 year tuition-free Founding Class Fellowship.

468K views across 2 posts

Theo called Grok 4.7 a disappointing release: 30 to 80% less token efficient than promised, slower than Grok 4.6, and more than 2x its real-world cost. On the same day SpaceXAI described Grok Bot running its customer support after the Cursor merger, handling a 175% rise in tickets without new hires and resolving 99% of refund requests on its own.

405K views across 1 post

OpenAI committed to giving independent assessors deep access across training, evaluation and deployment, so they can challenge its assumptions and check its safeguards. It published 4 priority areas for these assessments and principles for keeping the work rigorous, secure and independent.

274K views across 4 posts

Several posts described the same setup for coding agents: an orchestrator that plans and reviews while sub-agents write the code. Brad Groux has GPT-6 Astra delegate to 3 to 5 Sol threads, Vox shows Astra driving up to 10 Claude Code workers through Herdr, and Arpit Bhayani explains the "chief of staff" pattern as a way to keep each agent's context small.

129K views across 1 post

Lenny Rachitsky noted that nearly half of the 25 PM openings he shared last week ask for experience writing evals, with examples from Ramp, Shopify, Harvey and Cursor of what evals improved. He published an advanced evals guide by Hamel Husain and Shreya Shankar, with a free plugin that lets a coding agent do most of the work.