OUR THOUGHTS

On September 22, Anthropic cut Claude Opus 5.5's price by roughly 40 percent, and hours later OpenAI answered with GPT-6 Sol and Luna at about half of what their predecessors cost. Both cuts targeted agent work specifically: cheap enough to run all day, not just query once in a while. The next morning, Amazon showed what that buys. It opened Seller Central to outside agents, including Claude, and let independent sellers, who move more than 60 percent of everything sold on Amazon, approve an agent to set prices, manage inventory, and update listings on their own. For us, that's the real story, not the benchmarks. The price excuse for holding off on agentic AI expired this week, and the first company to notice didn't spend the savings on a smarter chatbot. It handed an agent the keys to a live business, with a single approval click as the only guardrail. The question worth bringing to your next pipeline review isn't which model is cheapest. It's which decisions in your own revenue motion you'd actually let an agent make unsupervised, and which ones you'd only trust it to draft.

This week we've curated three pieces that trace this in order. Anthropic's own Opus 5.5 launch and OpenAI's Sol and Luna launch, both from September 22, show this wasn't one lab moving alone: the market just repriced agentic AI as something you run all day, not something you query occasionally. GeekWire's reporting on Amazon's Seller Central APIs shows where that gets spent first: real operating authority over pricing and inventory, not more conversation. Together, they tell us the cost objection to agentic AI expired this week. For B2B sales and marketing leaders, that moves the real question up the calendar: not whether you can afford to run agents, but what yours are allowed to touch without you in the loop, and whether you've actually written that down.

ARTICLES OF THE WEEK

Introducing Claude Opus 5.5 – Anthropic
Published September 22, 2026, this article introduces Claude Opus 5.5 at 40 percent less cost to run than Opus 5 on typical workloads, alongside benchmark results Anthropic says put it ahead of OpenAI's GPT-6 Astra on coding and real-world task evaluations at a fraction of the cost per task.
Key takeaways:

• Opus 5.5 costs 40 percent less to run than Opus 5 on typical workloads, according to Anthropic, including a 60 percent cut on cache-read pricing, the input type that dominates spend on long-running agentic tasks.

• On Anthropic's GDPval-AA v2.1 evaluation, Opus 5.5 scored 1846 Elo against GPT-6 Astra's 1542, a result Anthropic says came at about a fifth of the cost per task.

• For B2B sales and marketing leaders, this closes the argument that frontier AI is too expensive to run continuously inside a revenue workflow; the remaining question is governance, not budget.
  Read more: https://www.anthropic.com/claude-opus-5-5

Introducing GPT-6 Sol and Luna – OpenAI
Published September 22, 2026, the same day as Anthropic's Opus 5.5 announcement, this article introduces two new task-specific GPT-6 models priced roughly 50 percent below their GPT-5.6 predecessors.
Key takeaways:

• Sol is built for complex, coding-heavy agentic work, while Luna targets high-volume, routine tasks, a deliberate split by job rather than a single general-purpose upgrade.

• OpenAI says that at its highest effort setting, Sol outperforms Claude Opus 5 at Opus 5's maximum effort on the AutomationBench evaluation at just 9 percent of Opus 5's cost per task.

• For revenue leaders comparing vendors, workload-specific pricing tiers mean the evaluation question is shifting from which model to run to which task belongs on which tier, a more granular decision than most GTM tech stacks are currently built to make.
  Read more: https://openai.com/index/introducing-gpt-6-sol-and-luna/

Amazon Opens Its Seller Tools to Outside AI Agents, Starting With Anthropic's Claude – GeekWire
Published September 23, 2026, this article reports that Amazon has opened its Seller Central APIs to outside AI agents in a US beta, letting independent sellers approve Anthropic's Claude or Amazon's own Quick assistant to manage inventory, pricing, and listings on their behalf.
Key takeaways:

• Sellers choose which data the plugin can reach and approve each action before it is carried out, an explicit human-in-the-loop design built into the rollout, according to GeekWire's reporting.

• Independent sellers already account for more than 60 percent of units sold on Amazon worldwide, and Amazon says about 90 percent of them already use some form of outside AI in their business, meaning this beta formalizes behavior that was already happening informally.

• For B2B leaders, this is a live example of what cheap, capable AI turns into within days of the price barrier falling: not another chat interface, but agents making pricing and inventory decisions inside a live revenue stream, exactly the kind of delegation most GTM organizations have not yet written rules for.
  Read more: https://www.geekwire.com/2026/amazon-opens-its-seller-tools-to-outside-ai-agents-starting-with-anthropics-claude/