For two years the answer to “which AI should our team pay for?” has been boringly simple: pick a $20 seat and move on. This week that calculus quietly broke. Anthropic dropped Claude Fable 5 — the public version of its most powerful model — straight into the plans people already hold, and an open-weight challenger now matches it on the work small teams actually care about. The headline price didn’t change. The value behind it did.
What Actually Landed This Week
Fable 5 went generally available on 9 June, and the interesting part is the packaging. Through 22 June it is included in Pro, Max, Team and seat-based Enterprise plans at no extra cost; after that, continued use draws on usage credits. On the API its metered rate is roughly $10 per million input tokens and $50 per million output tokens. It is strong at software engineering, knowledge work and vision — exactly the agentic territory — with hard safety limits that fall back to a smaller model in high-risk domains.
Meanwhile the flat tiers have converged. ChatGPT Plus and Claude Pro both sit at $20 a month; the headline subscription is now a commodity. So if everyone charges $20, the only number that matters is what you get inside it — and how fast you hit a wall.
Enter the Open-Weight Workhorse
This is where MiniMax M3 changes the maths. It is open-weight and genuinely frontier-grade at agentic coding, which means its marginal cost per token isn’t set by a vendor’s price list — it’s set by your hardware. Run it on a machine you already own and the next million tokens cost you electricity, not a credit balance.
That reframes the whole question. A flat $20 seat is a ceiling: generous until you’re mid-project, then suddenly rationed. An open-weight model you host is a floor: slower to set up, but the volume is yours.
$0marginal cost per token once an open-weight model like MiniMax M3 runs on hardware you already own.
The $19 Stack
Here’s the thesis, and the “$19” is deliberate — it’s a notch under the $20 everyone defaults to. For about the price of a single incumbent seat, a small team can assemble more usable agentic capability than that seat alone provides:
- Hard problems → Fable 5, during the bundled window. While it’s free inside the $20 Pro/Max tier, use it for the genuinely difficult agentic and software tasks. After 22 June, meter it deliberately rather than leaving it running.
- Volume work → MiniMax M3, self-hosted or cheap API. Triage, extraction, drafting, classification — the high-token, repetitive jobs — go to the open-weight workhorse where tokens are effectively uncapped.
- The wallet stays roughly flat. One thoughtful ~$19–20 of spend, split across a premium model for depth and an open model for breadth, out-tokens a single flat subscription every time.
The point isn’t that any one product costs $19. It’s that value per pound has decoupled from the sticker price. The teams getting the most from AI in mid-2026 aren’t the ones paying the most — they’re the ones who stopped thinking in seats and started thinking in tokens.
What This Means for a Small UK Team
Don’t reflexively renew a second or third $20 seat. Audit where your tokens actually go: if most of your usage is high-volume, repetitive work, an open-weight model like MiniMax M3 will quietly out-value any flat subscription, and the data never leaves your building. Reserve the premium model — Fable 5 while it’s bundled, metered thereafter — for the work that genuinely needs the top of the range. Spend like a buyer of tokens, not a buyer of logos, and a $19-shaped budget will comfortably beat a $20 one.
