Skip to main content

Changelog

Follow new updates and improvements to allmates.ai.

New frontier models: Gemini 4, Claude Fable 5.2, GPT-6.2 and Grok 4.7

What a month. In a few weeks, Google, Anthropic and OpenAI each shipped new frontier models, and xAI's Grok 4.7 came to a new connection. They are all in allmates.ai already, ready for your Mates.

You have probably felt the pace too. We announced Claude Fable 5.1 in early September, GPT-6.1 Sol in early October and Gemini 4 Argon only two weeks ago. Their successors are already here. The labs keep speeding up, and we keep up with them, so your Mates always have the best models at hand. Here is what to try first.

Gemini 4 Hydrogen: fast, and made for agent workflows

Hydrogen is Google's new fast Gemini 4 model, and Google reports it 2.5x faster than Gemini 3.8 Flash. Google pitches it for everyday coding, research and data analysis, and as the worker model that runs the steps of a multi-step agent workflow. It has a 1M-token context window. Try it on the research or analysis Mate your team uses every day. Gemini Pro Latest and Allmates large already run on it. Available on the Google connections that your organization brings itself: Vertex AI, Gemini API and OpenRouter.

Gemini 4 Carbon: for the long, hard work

Carbon is Hydrogen's big sibling. Google pitches it for "sustained agentic workflows" and multi-step problems in software engineering, scientific research and cybersecurity. Google sees the two as a duo: Carbon coordinates, Hydrogen runs the steps. Give Carbon to the Mate that carries your longest and most complex projects. Available on the same Google connections: Vertex AI, Gemini API and OpenRouter.

Claude Fable 5.2: Anthropic's new Fable

Barely two and a half months after Fable 5.1, Anthropic is back with Fable 5.2, which it positions "for demanding reasoning and long-horizon agentic work". It reasons on every request and has a 1M-token context window. It is a strong pick for a Mate that must think a problem through over many steps: a strategy review, a complex analysis, a long project. Claude Fable Latest already runs on it. Available on the Anthropic, Vertex AI and OpenRouter connections.

GPT-6.2 Sol and GPT-6.2 Astra: OpenAI's new pair

Seven weeks after GPT-6.1 Sol, OpenAI ships a new pair, each in standard and Pro. OpenAI sums it up well: "Use GPT-6.2 Astra for demanding reasoning and coding tasks, or GPT-6.2 Sol for everyday coding and agent workflows." So Sol goes to the Mates your team calls all day, and Astra to the ones that take on the hardest questions. Both read text and images, with a context window of about 1M tokens. Good news if a Mate already uses GPT-6.1 Sol or GPT-6 Astra on the OpenAI or OpenRouter connection: it now runs on the 6.2 successor, with nothing to do. GPT Latest now runs on GPT-6.2 Sol. Available on the OpenAI, OpenRouter and Azure AI Foundry connections.

Grok 4.7, now on Vertex AI

Teams on Google Cloud, this one is for you. Its Vertex AI model page describes Grok 4.7 as "a high-capability model from xAI, built for coding and knowledge work" that "works longer on difficult tasks and checks its own work". A good fit for a coding Mate or a research Mate. Already on the xAI and OpenRouter connections, Grok 4.7 is now also on the Vertex AI connection, where Grok Latest now runs on it.

Also in this release

  • Latest models and tiers follow on their own: Gemini Pro Latest and Allmates large run on Gemini 4 Hydrogen, Claude Fable Latest on Claude Fable 5.2, GPT Latest on GPT-6.2 Sol, and Grok Latest on the Vertex AI connection on Grok 4.7. Your Mates keep their setting and get the new model with no action from you.

  • Prices: Google lists Gemini 4 Hydrogen at 1.50 dollars per million input tokens and 7.50 dollars per million output tokens, against 4 and 20 for Gemini 4 Argon, which Allmates large used before. GPT-6.2 Sol keeps the price of GPT-6.1 Sol. GPT-6.2 Astra costs about 20% less than GPT-6 Astra, according to OpenAI's price list.

  • Deprecations: GPT-5.6 Sol is deprecated in favour of GPT-6.2 Sol. On Azure AI Foundry, GPT-6.2 Sol and Astra are added and GPT-6.1 Sol and GPT-6 Astra are marked deprecated. On Vertex AI, Grok 4.6 is marked deprecated.

  • Muse Ember (Preview): Meta's Muse Ember is here in preview on the OpenRouter connection.

  • Qwen4 Max: Qwen4 Max takes over from Qwen 3.8 Max on the OpenRouter connection.

Try one today

Pick one Mate and give it one of these models: open the Mate's settings, choose the model in the list, and see how it does on your real work. Every other Mate keeps its model until you change it. Learn more: Current lineup.

We can't wait to see what your Mates do with them.

— The allmates.ai Team

French public data, competitor prices and sixteen new connections

Sixteen connections joined the store over the last ten days. Here is what your Mates can now reach.

French public data

data.gouv.fr searches France's open public data — datasets, public APIs and their contents. IGN Geocontext looks up addresses, land parcels, zoning rules and administrative areas from IGN's official geographic data. Together they answer the questions behind a real-estate or site-selection file: what is this parcel, what can be built on it, and what does public data say about the area.

Competitor price watch for retailers

Six retailer listings give a Mate a competitor's online prices, with unit prices and barcodes to match products from one store to the next:

  • Grocery — Carrefour, E.Leclerc and Auchan (prices of the drive serving a postal code), plus French Grocery Prices for the three at once.

  • DIY — Leroy Merlin, Castorama and Brico Dépôt.

They run on your organization's own Apify account, and each listing tells your Mates how to search precisely: one product per search, a result limit, and products matched on their barcode.

Design, CRM and data governance

  • Figma — read designs, variables, components and screenshots, and write back to the canvas.

  • Salesforce — two listings: read-only, and read & write, always under each person's own Salesforce permissions.

  • DataGalaxy — search the data catalog and read the glossary, lineage and ownership of any asset.

Integration and language

  • Boomi Connect and Boomi API Control Plane — the enterprise apps and APIs your Boomi admin publishes.

  • DeepL — translate, rephrase and proofread text and documents with your DeepL account.

Zernio, rebuilt

Zernio now lists its official server's real tools: publish and schedule posts, answer comments and mentions, and read analytics across sixteen social platforms. Connect it with an organization API key, or let each person sign in with their own Zernio account — their Zernio role then applies.

New

🌌 The Mythos Era Begins: Claude Fable 5 is Now Live!

Anthropic has just released Claude Fable 5, the very first publicly available model from the highly anticipated Mythos family. Designed to tackle tasks of unprecedented complexity, Fable 5 represents a monumental shift in AI capability. It is available right now to power your Mates on allmates.ai.

⚠️ Crucial Cost Warning: Use with Precision

Claude Fable 5 is an ultra-premium, specialized powerhouse. It costs exactly twice as much as Claude Opus 4.8 ($10 per million input tokens / $50 per million output tokens).

To optimize your organization's credits, do not use Fable 5 for everyday, routine tasks. Reserve this model for highly complex, multi-step, or deeply analytical problems where standard models fall short. For daily operations, Claude Opus 4.8 or Claude Sonnet 4.6 remain your most efficient workhorses.

🔍 What’s New?

Claude Fable 5 brings Mythos-class intelligence out of the research lab and directly into your workflows.

  • 🧠 Mythos-Class Reasoning: Fable 5 delivers a significant leap in capability over Opus 4.8, scoring over 10% higher on advanced reasoning, scientific, and knowledge work benchmarks.

  • 👁️ State-of-the-Art Multimodal Vision: It features near-perfect visual accuracy. It can parse dense tables, extract precise data from complex charts, and even rebuild code structures from screenshots alone.

  • 📚 1-Million-Token Context & 128k Output: Process massive document archives, multi-session projects, or complete codebases in a single, uninterrupted turn.

  • 🛡️ Stricter Safety & Automatic Fallback: Fable 5 is built with advanced safety guardrails. If a query touches high-risk domains (like cybersecurity), the system will automatically and safely route the request to Claude Opus 4.8.

✨ Why You’ll Love It

  • Solve the Unsolvable: Perfect for deep-dive financial forecasting, massive document synthesis, and long-horizon strategic planning that typically requires hours of human oversight.

  • Enterprise-Proven Power: In early testing, companies reported compressing months of complex analytical work into mere days using this new architecture.

  • Flexible Deployment: Run it on your terms. Connect your own keys to maintain full control over your billing and quotas.

🛠️ How to Get Started (BYOK)

Because of its premium nature, Claude Fable 5 is available on allmates.ai via your own configured providers:

  1. Navigate to your Mate's settings on allmates.ai.

  2. Ensure you have active connections set up for Anthropic, Google Vertex AI, or OpenRouter.

  3. Select Claude Fable 5 or Latest from the model dropdown list.

We are thrilled to let you explore the absolute frontier of AI collaboration. Use this power wisely!

Happy collaborating,


— The allmates.ai Team

New

Earlier updates