Anthropic and OpenAI cut prices on the same day

For decades, farmers have relied on herbicides and heavy equipment to control weeds.

Greenfield Robotics is working to change that.

BOTONY robots navigate crop rows and mechanically remove weeds, giving farmers an alternative to blanket herbicide applications and intensive tillage.

The potential benefits are significant:

And the market is enormous.

Greenfield estimates more than 250 million acres of broadacre cropland in the U.S., with more than 100 million acres in its initial serviceable market.

With the Reg A+ now live, investors can participate in Greenfield Robotics' growth story.

Explore the Investment

^{This Reg A+ offering is made available through StartEngine Primary, LLC, member FINRA/SIPC. Please read the Offering Circular and related disclosures before investing. This investment is speculative, illiquid, and involves a high degree of risk, including the possible loss of your entire investment.}

1.  Anthropic ships Claude Opus 5.5 at 40% lower cost. Anthropic released Opus 5.5, which it positions as near-Fable performance for less money: $4 per million input tokens against Opus 5's $5, $20 output against $25, and cached reads down 60%. On Terminal-Bench 4.0 it scores 66.4% versus Opus 5's 52.3%, and one tester ran a 680,000-line code migration in under a day. The change users will notice first is the writing, which is tighter and less padded. Five-hour usage limits went up on every paid plan, and subscribers get a rate-limit reset they can bank for when they need it. A prompting guide is up if you want to tune your prompts.

2.  OpenAI answers with GPT-6 Sol and Luna. OpenAI introduced two cheaper GPT-6 tiers. Sol is for demanding professional work at $2 per million input and $10 output; Luna handles everyday tasks at $0.10 and $0.50. OpenAI calls that a 50% cut against the previous generation and says Sol matches Claude Opus 5 on computer-use tasks at about 80% lower cost per task. Both are in ChatGPT, Codex and the API for paid plans; Luna is free in the desktop app. Astra remains the top model.

3.  Mirage gives agents a real video editor. Mirage launched Tesseract, a suite that lets an AI agent work inside a proper editing environment, with compositions, keyframes and audio, while you watch a live preview and render on your own machine. It connects to ChatGPT and Claude, so the assistant you already use can now cut a video rather than write a script for one.

4.  Pocket FM builds a model for long stories. Pocket FM released Sherpa, which it calls a Narrative World Model, built to plan and hold together a story arc across many chapters rather than generate one scene at a time. The launch video passed 6M views. It is the first frontier-scale model aimed at serialized fiction, which is a bigger market than most of tech realizes.

The setup. On September 12, Dario Amodei asked the industry to pace itself and Sam Altman agreed. On September 23 both companies shipped a new model and cut prices, and neither announcement mentioned pacing. That is not hypocrisy; pacing was always about how fast capabilities grow, not how cheaply they are sold. But the two launches, read together, show where the actual competition has moved.

The pincer. Look at the price ladders. Anthropic's Opus 5.5 comes in at $4 in and $20 out, pitched as nearly Fable-grade. OpenAI's Sol sits at $2 and $10 and claims Opus 5 performance at a fifth of the cost; Luna at 10 cents and 50 cents undercuts everything. Anthropic is squeezing the top of the market from above, offering more capability per dollar; OpenAI is squeezing from below, offering enough capability for far fewer dollars. The frontier tier at each company, Fable and Astra, is untouched.

The tell. Both companies raised usage limits on the same day, and Anthropic added a reset users can save. You do not loosen limits when compute is scarce. You loosen them when you have capacity to fill and a rival is about to take your customers. The inference cost curve is falling faster than either company is advertising, and the price cuts are passing some of it along.

What it means for you. The practical change is that the model you were rationing last month is now cheap enough to use freely. If you are on a subscription, your limits went up. If you pay by the token, re-check which tier you actually need: Luna will handle most drafting and summarizing, Sol most analysis, and the top models are for the work where a wrong answer is expensive. Routing by task is now worth an hour of setup.

The reframe. Slowing capabilities and racing on price are not in tension; they may be the same strategy. A company that expects capability growth to be paced has one lever left to win share, and that is making its current model the cheapest good option. Watch what happens to the frontier tiers over the next month. If Fable and Astra get cheaper too, the pacing pledge is holding. If they get more capable, it is not.

The Opus playbook. Claude's developer team published a guide to getting the most out of Opus 5.5, from prompting to steering mid-task to checking results, and it passed 5,000 bookmarks in two hours. If you upgraded today, read this before you change your prompts.

The one-button hedge fund. Bracket22, a fund run entirely by AI agents, went viral after a CNBC clip in which manager Brian Kelly said his job now consists mostly of occasionally pressing a button. Returns are unverified; the picture of a one-person firm is the point.

Muse fights your airline. A Muse user posted that Meta's agent won them a $250 flight-delay credit without their involvement and described the whole thing as magic. The claim is a single post on Superhuman's archive, so treat it as an anecdote, but it is exactly the use case Meta is selling.

Grok Bot for onboarding. The workflow making the rounds this week: a Grok Bot connected to your HR documents that builds a new hire's first-week checklist, answers their questions from those documents only, and flags anything that needs a human. Keep the no-changes-without-approval rule in the brief.

Three tools worth a look. Clueso makes polished product videos from a screen recording, WZRD gives you documents and slides you can edit by talking to them, and thestory.run coaches people who post professionally to sound like themselves.

Turn one article into a week of social posts

You wrote the piece, or the report, or the transcript exists. Getting it onto LinkedIn and X without repeating yourself five times is the part nobody has time for. This prompt hands the source to ChatGPT or Claude and gets back a set of standalone posts, each built on a different angle from the same material, with the hook and caption separated so you can swap one without rewriting the other.

Set [NUMBER] to five for a week of daily posts, and pick one platform per run; LinkedIn and X want different lengths and the prompt adapts to whichever you name. Advanced tip: paste this newsletter as the source and ask for the contrarian angle only, then compare it to what you would have written.