Zapier's new model matches Opus 5.5 at a third of the price
Zapier confirmed GPT-6.1 Sol matches Opus 5.5 on its own workflow benchmark at about a third of the cost.
Four-minute read. In today’s issue: Zapier’s new default model matches Opus 5.5 for a third of the price, OpenAI’s six-times-pricier Ultrafast mode, Gemini skills replacing Gems, and a Terminator-style ending for some robots.
PRICING
GPT-6.1 Sol runs Zapier workflows at a third of Opus 5.5’s cost
What changed. Zapier confirmed GPT-6.1 Sol is now live inside its automation platform, and Zapier CEO Wade Foster said on X that the model matches Opus 5.5 on Zapier’s own multi-step business-workflow benchmark, AutomationBench, which covers sales, marketing, ops, support, finance and HR tasks, at roughly a third of the price.
What it means at your size. If you run Zaps that hand off work between sales, support or finance steps, you may be able to switch the model behind them without losing the reasoning those steps need, and cut what each run costs you.
The first step. Open your Zapier account this week and check which model your existing Zaps are set to use; flag the ones on a pricier model so you can test GPT-6.1 Sol against them before committing.
The limitation. The price claim is Foster’s own read of Zapier’s internal benchmark, not an independent test, and the briefing gives no dollar figures beyond “roughly a third” of Opus 5.5’s cost.
Wade Foster on GPT-6.1 Sol matching Opus 5.5 at a third the price (x.com)
SPEED
OpenAI’s new Ultrafast mode costs six times as much
OpenAI launched Ultrafast, a paid speed setting for GPT-6 Astra that generates up to 300 tokens a second, about six times faster than normal.
The catch is the price: $60 per million input tokens and $300 per million output. That’s exactly six times standard Astra, and 30 times GPT-6.1 Sol, which launched the same day.
Amazon Web Services is already reselling the same tier on Bedrock at the same price, or 10% more if you keep your data in the US.
Reactions split between “impressive” and “outrageous”, with The Decoder noting the 6x markup applies to every line of the price sheet.
For a small team, speed is worth paying for only when someone sits waiting on every reply, like a live chat agent. For email drafts, reports and overnight automations, it buys you nothing, and Sol does the job for a fraction of the cost. Your ChatGPT or Claude subscription isn’t affected; this pricing is for API use only.
OpenAI on the new Ultrafast mode · AWS on Ultrafast on Bedrock
🚀 Just shipped
Google Workspace Updates says Google Vids now lets you customize caption styling, live today for anyone building videos in Vids. If you use Vids for training clips or social posts, you can match captions to your brand without a separate editing tool. (workspaceupdates.googleblog.com)
📌 The rest of the day
🌐 Google DeepMind launched Gemini 4 Argon, a new frontier model pitched at complex work, enterprise knowledge tasks and cyber defense, rolling out first to trusted testers under its “Fairwind Program” rather than to everyone at once. (deepmind.google)
🧩 Google is adding reusable “skills” to Gemini and Workspace from October 5, so a small business can save its brand voice once and apply it across Docs and Gmail. Skills replace Gems, which stop working on business accounts in March 2027. (workspaceupdates.googleblog.com)
🔗 OpenAI’s developer account said shareable profiles now bring a builder’s Sites and plugins together in one link inside ChatGPT, so a small business running a custom GPT or plugin can point customers to a single page instead of scattered links. (x.com)
💬 Peter Yang and Nate Herk both spent the week putting OpenAI’s Dots up against Meta’s Muse, the two leading personal-agent apps, with Yang arguing the pair are settling into separate lanes: one for work tasks, one for personal ones. (x.com)
🛠️ Built with AI
Thariq, who works at Anthropic, had Opus 5.5 run through a series of redesigns of his personal website, critiquing its own drafts along the way, then cut a trailer of the whole process; you could run the same loop on your own homepage and come away with both a new design and a highlight reel to post. (x.com)
🔭 Too early
Innate, a seven-person robotics startup, posted that it’s building “capable robots for everyday life.” There’s no product, price or date attached yet, so it’s one to watch rather than plan around. (x.com)
🎮 One for the road
Figure needed to retire its old F.02 humanoid robots, and taking them apart by hand would have slowed down the next model.
When it asked online for ideas, Arnold Schwarzenegger suggested melting them down, Terminator-style, then joined in.
Figure trained the robots on stunt performers’ moves and practised jumps off the second floor of its San Jose office. Then it shipped them to a foundry in Imatra, Finland, the only one in the world that would take them.
There, the robots jumped on their own into buckets of molten steel: six melts, each with about 20 minutes before the steel cooled.
The bars went back to the US and became limited-edition collectibles, so you can buy a piece of a retired robot in Figure’s shop.
Yes, the footage is real. Some retirement plan.