Wednesday’s through-line is cheaper, more capable AI landing in tools people already use. Anthropic shipped Claude Opus 5.5 at a lower run cost. OpenAI answered with GPT-6 Sol and Luna. Amazon rolled out free agentic “workflows” for marketplace sellers. And Meta’s Muse agent is climbing the App Store as Connect kicks off.
1. Anthropic’s Claude Opus 5.5: frontier work for less
On September 22, 2026, Anthropic introduced Claude Opus 5.5, the first model in its Claude 5.5 family. Anthropic says it performs at the level of Claude Fable 5.1 on most work and costs about 40% less to run than Opus 5 on typical workloads. API pricing is $4 per million input tokens and $20 per million output tokens, with cache reads at $0.20 per million (a big deal for long coding and agent sessions).
The company positions Opus 5.5 as stronger on agentic coding, computer use, and knowledge work, and says it scored best so far on Anthropic’s automated behavioral audit. It ships with safeguards similar to Fable 5.1 for biology and cybersecurity, plus “preserved thinking” for newer API accounts. Opus 5.5 is available across Anthropic’s platform and major clouds (AWS, Google Cloud, Azure). Sonnet 5.5 and Haiku 5.5 are promised in the coming weeks.
Why it matters: if you pay for Claude API use, or you run long Claude Code / agent sessions, a cheaper Opus-class model is a real budget and speed change, not just a benchmark flex. Re-check which model your tools default to, and compare cost per finished task (not only sticker price per token).
Sources: Anthropic – Claude Opus 5.5.
2. OpenAI’s GPT-6 Sol and Luna: Astra-class help at lower price points
Hours apart from Anthropic’s drop, OpenAI rolled out GPT-6 Sol and GPT-6 Luna. They sit under GPT-6 Astra and aim to bring much of that stack into faster, more affordable models. OpenAI says API prices are 50% lower than GPT-5.6 promotional pricing, with higher usage limits and a banked rate-limit reset for Plus, Pro, and Business accounts.
Official docs list Sol at $2 / $10 per million input/output tokens (built for complex coding and agentic workflows) and Luna at $0.10 / $0.50 (aimed at focused, high-volume tasks). Both advertise a 1.05 million token context window. Sol and Luna are in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu; both are in the API; Free and Go users can try Luna in the desktop app.
Why it matters: small teams that stayed on older GPT tiers for cost can try GPT-6-class work without jumping straight to Astra pricing. Pick Luna for high-volume drafts and triage, Sol when the job needs heavier coding or multi-step agents, and keep a human check on anything that spends money or changes production systems.
Sources: OpenAI Developer Community – Announcing GPT-6 Sol and Luna, OpenAI – GPT-6 Sol, OpenAI – GPT-6 Luna.
3. Amazon’s free seller “workflows” that watch ratings and prices
On September 23, 2026, Reuters reported that Amazon launched a new agentic AI service for third-party sellers on its main store. The feature is called workflows: continuous prompts that can alert you if ratings suddenly drop, or watch pricing for specific products. The Seller Assistant can store a seller profile for recommendations on pricing, promotions, and inventory. It can also plug into tools sellers already use, starting with Amazon’s Quick and Anthropic’s Claude. Amazon’s Mary Beth Westmoreland said the agent is free and optional, and sellers control how much personal data they share.
Why it matters: if you sell on Amazon (or you advise someone who does), this is the kind of agent that can save hours of dashboard babysitting. Treat it like a junior ops hire: useful for alerts and drafts, not a blank check to change prices or inventory without your review. Start with monitoring workflows before you let it act.
Sources: Reuters – Amazon rolls out new agentic AI for third-party sellers.
4. Meta’s Muse agent is popular. Trust is still the product question.
Also on September 23, CNN Business published a hands-on look at Meta’s Muse AI agent (launched September 8). Sensor Tower data cited by CNN put downloads above 2.5 million, with Muse as the most popular free iPhone app at the time of reporting. Muse can plan packing, draft emails, build trip docs, and browse the web on a cloud virtual computer you can watch and interrupt. Mark Zuckerberg is expected to talk more about Muse at Meta Connect on Wednesday.
The same piece flags the limits: stale local recommendations, shopping sites that block bot checkout, and prompt ideas that oversell what Muse can do without you (for example, Marketplace haggling that still needs you to send the message). Privacy and trust remain the adoption hurdle, especially for an agent that wants calendar, email, and payment-adjacent access.
Why it matters: personal agents are moving from demos to App Store defaults. If you try Muse (or any peer), connect the minimum accounts first, approve purchases yourself, and treat “it can do it for you” as “it can draft and navigate until a human confirms.”
Sources: CNN Business – Meta’s Muse AI agent hands-on.
What to do with Wednesday’s news
- Paying for Claude or ChatGPT API use? Benchmark Opus 5.5, GPT-6 Sol, and Luna on one real workflow you already run. Compare cost and cleanup time, not just demos.
- Selling on Amazon? Turn on optional monitoring workflows for ratings and price watches before you grant action rights.
- Curious about Muse or similar agents? Give them planning and drafting first. Keep payments, messages to strangers, and account changes on a human click.
That is Today in AI for September 23, 2026: cheaper frontier models from Anthropic and OpenAI, free seller agents on Amazon, and a Meta personal agent that still needs your judgment.
Share this article
Leave a Reply