Spojit
Newsletter

September review: Opus 5.5, Fable 5.1, and Miraxa on Auto

September was a busy month for new models. Anthropic shipped two, Google shipped one, and all three are on Spojit now. The headline is Claude Opus 5.5: it is our new default, and it costs fewer credits than the model it replaces.

Miraxa also learned to finish a job without stopping at every step, and connecting a vendor's own MCP server now takes a URL and a click. Here is the tour.

Model update

Claude Opus 5.5: the new default, for fewer credits

Anthropic's newest Opus performs at the level of Claude Fable 5.1 on most work, and ahead of it on Anthropic's agentic coding and knowledge work benchmarks. It also writes its answers more than 30% faster than Opus 5.

And it is cheaper. Every token costs 20% fewer credits than on Opus 5, so the better model is also the smaller bill.

Miraxa now uses it by default on every plan, and so do new workflows on paid plans. Existing workflows keep whatever model you picked for them, and Opus 5 stays available if you want it.

See the full model guide
Model update

Claude Fable 5.1: for when a wrong answer is expensive

Fable is Anthropic's flagship tier, built for the deepest reasoning and long autonomous runs, and 5.1 is a big step on Fable 5: it more than doubled its score on Anthropic's Terminal-Bench-Science, from 24.7% to 52.6%.

It costs the same credits as Fable 5 and works in Miraxa and every workflow step. With Opus 5.5 this close, start on Opus and save Fable 5.1 for the problems where the extra margin is worth paying for.

Model update

Gemini 3.8 Flash, the most capable Flash yet

Google's newest Flash is in your workflow steps now, built for long multi-step agent work. Here is how the current Gemini lineup splits:

  • Gemini 3.8 Flash the most capable Flash, for steps where the reasoning counts
  • Gemini 3.6 Flash quicker than 3.8 for the same credits
  • Gemini 3.5 Flash-Lite fastest and fewest credits, built for high volume
Compare the Gemini models
Action needed

Gemini 2.5 retires on 20 October

Google retires Gemini 2.5 Pro and Gemini 2.5 Flash on 20 October 2026. Any workflow step still set to either will stop working that day, with a message saying why. Switch those steps to Gemini 3.8 Flash or 3.5 Flash before then.

Miraxa

Ask when you want a say. Auto when you don't.

Last month we introduced Operator mode: Miraxa working on your live systems and checking with you before each change. That mode is now called Ask, it is the default, and it is no longer limited to certain roles. If you can use Miraxa, you can use Ask.

Alongside it is Auto. Same Miraxa, same systems, but no pause before every change. Your governance policy still sets the limits: anything it marks for approval still waits for a person, anything destructive such as a delete still pauses, and anything it blocks stays blocked.

A good rhythm is to work out the approach in Plan, run it once in Ask, and hand the repeats to Auto. Auto is there for Editors, Admins and Owners, and an administrator can grant it to other roles. Miraxa remembers the mode you chose last.

How Ask and Auto work
Connections

Connect a vendor's MCP server with a URL and a click

More and more vendors host their own MCP server: Notion, Linear, Atlassian, NetSuite and plenty more. Connecting one used to mean setting up an OAuth app in the vendor's developer console first.

Not any more. Paste the server URL, leave authentication on Auto, click Save & connect, and approve access on the vendor's own screen. Spojit works out how the server signs in and registers itself.

To go with it, there are now step-by-step guides for over 30 popular servers, each with the exact URL, what you need on the vendor's side, and how to sign in.

Browse the MCP server guides
Also this month

Smaller things you might have missed

  • Miraxa docks beside your work pin it right, left or bottom, or pop it into its own window. Open conversations are tabs that follow you between devices.
  • Scanned PDFs in Knowledge pages without a text layer are now read with OCR, up to 100 scanned pages per document.
  • OAuth 2.0 on HTTP Requests call an API that signs in with OAuth by bringing your own OAuth app to the HTTP connector.
  • Loops that keep going set a loop to continue with the next item, so one bad row no longer stops an import, and act on the failures afterwards.
Try it

It is already there

Everything here is live now, with nothing to switch on. If you have workflows on Gemini 2.5, that is the one to do this week.

Product news

Get the next issue

New models, new platform capability and the releases behind them. Occasional, never noisy.

We will email you a link to confirm. Unsubscribe any time. See our privacy policy.