New on CompanyFabric: Claude Fable 5.2, Gemini 4 Argon, Muse Wonder, MiMo V2.6, Kimi K3.1 and 3 Decision Models

CompanyFabric is one API key for every model. It works with any OpenAI SDK: change the base URL, keep your code. Prices are listed per million tokens, exactly as the providers charge them, plus a small, published fee.

In the last few weeks we added five new frontier model families and a new kind of model: decision models, which return typed answers with probabilities instead of text. Here is what's new and how to use each one.

One setup for every model below

Every chat model in this post uses the same code. Only the model string changes.

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.companyfabric.com/v1",
    api_key=os.environ["COMPANYFABRIC_API_KEY"],
)

reply = client.chat.completions.create(
    model="anthropic/claude-fable-5-2",  # swap for any model id in this post
    messages=[{"role": "user", "content": "Plan a migration from REST to GraphQL."}],
)
print(reply.choices[0].message.content)

Get a key at companyfabric.com/dashboard/keys. New accounts get a welcome credit.

1. Anthropic: Claude Fable 5.2 and the Claude 5.5 family

Claude Fable 5.2 (released Nov 9, 2026) is Anthropic's most capable generally available model, built for demanding reasoning and long-horizon agentic work. It costs $8 input / $40 output per 1M tokens, 20% below Fable 5.1. It has a 1M-token context window and 128K output, and cache reads cost $0.20 per 1M tokens.

• Use it for: multi-hour coding agents, research that spans many documents, and hard planning problems where a cheaper model's evals still fall short.
• Model id: anthropic/claude-fable-5-2
• Learn more: Claude Fable 5.2 API: pricing, context and code samples

The rest of the Claude 5.5 generation is also live:

Using Claude Code? CompanyFabric also speaks Anthropic's own /v1/messages format. Point Claude Code at it and prompt caching works and is billed at the cache rate.

2. Google: Gemini 4 Argon

Gemini 4 Argon is Google DeepMind's frontier model for long-horizon coding, legal, finance and science work. It ranked #1 on the Vals Index at launch. Until November 30 it has an introductory price of $2 / $10 per 1M tokens. After that it moves to Google's standard $4 / $20.

• Use it for: contract review, financial analysis and long-context coding tasks with a 1M-token window.
• Model id: google/gemini-4-argon
• Learn more: Gemini 4 Argon API

3. Meta: Muse Wonder

Muse Wonder is Meta's first frontier model family. It is built for long-running personal and professional agent work and served by Meta directly. It costs $4 / $20 per 1M tokens, has a 1M context window, and supports tools, vision and prompt caching.

• Use it for: assistants that run for many steps, such as booking, research or inbox triage. It is also a strong second opinion next to Claude and Gemini.
• Model id: meta/muse-wonder
• Learn more: Muse Wonder API

4. Xiaomi: MiMo V2.6 Pro Ultraspeed and Flash

Xiaomi's MiMo line added two speed tiers:

• MiMo-V2.6-Pro-Ultraspeed: MiMo-V2.6-Pro's flagship quality with output up to 20x faster, for real-time, latency-sensitive work. $4.35 / $8.70 per 1M tokens. Model id xiaomi/mimo-v2-6-pro-ultraspeed. Learn more
• MiMo-V2.6-Flash: a low-cost, full-modality reasoning model for high-frequency calls, at just $0.14 / $0.28 per 1M tokens. Model id xiaomi/mimo-v2-6-flash. Learn more

Both have a 1M-token context window.

5. Moonshot AI: Kimi K3.1 Pro and Highspeed

Kimi K3.1 Pro is Moonshot's 3T-class open-weight flagship. It has a 1M context window and is built for coding, 3D and financial agents. It costs $2 / $10 per 1M tokens, with cache reads at $0.20.

Kimi K3.1 Pro Highspeed runs the same model 2–3x faster for responsive, interactive agents, at $4 / $20.

• Model ids: moonshotai/kimi-k3.1-pro, moonshotai/kimi-k3.1-pro-highspeed
• Learn more: Kimi K3.1 Pro API

New: decision models (Jev, Perplexity Decider 27B, Microsoft-Decision-1)

Chat models write text that you then have to parse. A decision model never writes text. You send it the situation (a support ticket, a brief, a transcript, a JSON record) and a set of typed questions. It returns one typed answer per question, each with a calibrated probability, in a single fast pass. There is no JSON to validate and no free text to misread, and it can't invent an answer outside the options you gave it.

There are three question types:

• noul: a yes/no probability ("Does this need a reply within the hour?")
• choice: picks one label from criteria you describe ("billing", "technical", "other")
• score: places the input on a rubric you define ("no signal" → "threatens to cancel")

All three models take the exact same request on POST /v1/decisions, so you can compare them on your own data by changing one field. Output is free on all three, so a typical call costs a few thousandths of a cent.

curl https://api.companyfabric.com/v1/decisions \
  -H "Authorization: Bearer $COMPANYFABRIC_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "typesafe/jev-latest",
    "state": "Second time this month my card was charged twice. Refund me or I cancel.",
    "questions": {
      "category": {
        "type": "choice",
        "instructions": "What is this ticket about?",
        "criteria": { "billing": "Charges, refunds", "technical": "Bugs, errors", "other": null }
      },
      "urgent": { "type": "noul", "instructions": "Does this need a reply within the hour?" },
      "churn_risk": {
        "type": "score",
        "instructions": "How likely is this customer to leave?",
        "criteria": ["No signal", "Frustrated", "Threatens to cancel"]
      }
    }
  }'

The response comes back with probabilities you can route on directly:

{
  "model": "typesafe/jev-latest",
  "answers": {
    "category":   { "type": "choice", "choice": "billing", "confidence": 0.97 },
    "urgent":     { "type": "noul", "noul": 0.88 },
    "churn_risk": { "type": "score", "score": 1.8, "confidence": 0.81 }
  },
  "usage": { "input_tokens": 142, "output_tokens": 0, "cost_usd": 0.0000063 }
}

Jev by TypeSafe

Jev returns typed decisions with calibrated probabilities in about 100 ms. It is the default when you omit model. $0.042 per 1M input tokens, output free.Model id typesafe/jev-latest · Try Jev in the playground

Perplexity Decider 27B

Perplexity's decision model works the same way, and it can also read images in state. Use it for screenshots, receipts or product photos. $0.04 per 1M input tokens, output free.Model id perplexity/decider-27b · Try Decider 27B

Microsoft-Decision-1

Microsoft's decision model launched in Microsoft Foundry on October 9. In Microsoft's own 36-benchmark comparison it ranked first on accuracy and was the fastest model measured. $0.042 per 1M input tokens, output free.Model id microsoft/microsoft-decision-1 · Try Microsoft-Decision-1

Where decision models fit: ticket triage, lead scoring, content moderation, routing an agent's next step, and eval grading. They work anywhere you currently ask a chat model to "answer with JSON" and then hope it does. Full reference: Decisions API docs.

Get started

1. Create a key at companyfabric.com/dashboard/keys. The welcome credit covers your first calls.
2. Set base_url to https://api.companyfabric.com/v1 in any OpenAI SDK.
3. Pick a model id from this post, or use companyfabric/auto and let the router choose the cheapest model that fits the request.

See everything we've shipped on the CompanyFabric changelog.