What's new
in CompanyFabric.
New models, endpoints, dashboard and billing changes, newest first. Models land here the day their provider ships them.
Last updated · 50 updates
November 2026
7 updates
- Model
Gemini 4 Hydrogen
Google's fast, efficient Gemini 4 model for everyday coding and agentic work, live the day it launched: $1.50 input and $7.50 output per million tokens, flat across its 1M-token window.
See it: Gemini 4 Hydrogen - Model
Gemini 4 Carbon is live
Google's strongest reasoning model, announced on 28 October, can now be called on the same key and balance. It spends more thinking tokens than Argon to verify its work, at the same $2 input and $10 output per million tokens until 30 November.
See it: Gemini 4 Carbon is live - Model
Gemini Decision 1
Google's decision model, live the day it launched: typed yes/no, choice and score questions about text, images, audio, video or PDFs, answered with probabilities for $0.10 per million input tokens.
See it: Gemini Decision 1 - Model
Claude Fable 5.2
Anthropic's most capable generally available Claude is live on the same key and balance, priced 20% below Fable 5.1.
See it: Claude Fable 5.2 - Site
Model library as a price list
Every model is now one row with context, input price, output price and the date it went live, so you can compare prices down a column instead of opening cards.
See it: Model library as a price list - Dashboard
Clearer usage charts
Daily spend, requests and tokens now have real axes and tooltips, and failed requests are drawn on top of successful ones so an error spike is visible at a glance.
- Site
Agent templates
Open-source starter apps you can clone and run on your CompanyFabric key, each with screenshots, the source on GitHub and an install command.
See it: Agent templates
October 2026
18 updates
- Dashboard
Scopes on API keys
Choose what each key is allowed to do when you create it, so a key you hand to one tool cannot be used for everything on your account.
- Model
Gemini 4 Carbon, tracked before launch
Google has announced it but not opened the API. Its page carries the published launch price and will switch to live the day it can be called.
See it: Gemini 4 Carbon, tracked before launch - API
Web search: POST /v1/search
Search the web with the same key and get results billed per request. It is also available to agents as a web_search tool on the MCP server.
See it: Web search: POST /v1/search - API
OpenAI Responses API: POST /v1/responses
Code that calls client.responses.create, or the Vercel AI SDK's default OpenAI provider, now works unchanged against every model here.
See it: OpenAI Responses API: POST /v1/responses - Model
Kimi K3.1 Pro and K3.1 Pro Highspeed
Moonshot's open-weight coding flagship with a 1M context window, plus a faster tier for interactive agents at twice the price.
See it: Kimi K3.1 Pro and K3.1 Pro Highspeed - API
Anthropic-compatible /v1/messages
Claude Code and other tools that speak Anthropic's own format can point at CompanyFabric directly, with Claude tool use and cache pricing supported.
See it: Anthropic-compatible /v1/messages - Dashboard
See which model answered an auto request
When you call companyfabric/auto, the usage page now names the model that actually served each request, so the price you paid is explained.
- API
Long requests get up to 800 seconds
Chat completions and messages requests can now run for as long as a deep reasoning model needs before the gateway gives up.
- Model
MiMo-V2.6-Flash and MiMo-V2.6-Pro-Ultraspeed
Xiaomi's low-cost reasoning model for high-volume calls, and a tier of its flagship with much faster output for real-time use.
See it: MiMo-V2.6-Flash and MiMo-V2.6-Pro-Ultraspeed - Model
Gemini 4 Argon is live
Tracked here since Google announced it on 30 September, and callable from the day the API opened.
See it: Gemini 4 Argon is live - Model
- Model
Microsoft-Decision-1
A third decision model on /v1/decisions: ask typed questions and get calibrated probabilities back instead of text to parse.
See it: Microsoft-Decision-1 - Model
Claude Haiku 5.5, and models listed newest first
Anthropic's cheapest and fastest Claude is live. Every model list on the site now runs newest first with the date each one arrived.
See it: Claude Haiku 5.5, and models listed newest first - Model
Mistral Large 4
Mistral's open-weight flagship, available here from its public preview.
See it: Mistral Large 4 - Dashboard
Low-credit alert
The dashboard warns you when your balance is running down, before requests start failing.
- API
OpenAPI spec at /openapi.json
A machine-readable description of the API, for generating clients or giving an agent the full contract in one file.
See it: OpenAPI spec at /openapi.json - Model
Perplexity Decider 27B
A second decision model next to Jev, with its own page, playground and starter examples.
See it: Perplexity Decider 27B - Chat
Retry one model without re-running the others
If one model in a comparison fails, retry just that column. A reply that died now says so instead of showing "Thinking" forever.
See it: Retry one model without re-running the others
September 2026
18 updates
- Model
Claude Sonnet 5.5
Near-Opus coding and knowledge work at Sonnet prices, live on launch.
See it: Claude Sonnet 5.5 - Dashboard
Budget caps and spend per API key
Set a monthly limit on any key and see what each key has spent, so one runaway script cannot drain the whole balance.
- Billing
Prompt-cache hits billed at the cache rate
Input the provider serves from its cache is charged at the provider's lower cache price, and the usage page shows how much that saved you.
- Dashboard
Every request, listed
The usage page lists each request with its model, tokens and exact cost, alongside daily tokens split into cached input, fresh input and output.
- Dashboard
Evals
Run one agent setup against up to five models and get each answer graded per check, so choosing a model is a measurement instead of a guess.
- API
Decisions: POST /v1/decisions
Send a situation and typed questions, get probabilities back. Launched with Jev, with a playground and starter examples.
See it: Decisions: POST /v1/decisions - Model
Claude Opus 5.5, Claude Fable 5.1, GPT-6 Sol, GPT-6 Luna, GLM-5.3
Five frontier models added in one day, each with verified prices.
See it: Claude Opus 5.5, Claude Fable 5.1, GPT-6 Sol, GPT-6 Luna, GLM-5.3 - Model
Xiaomi MiMo, Meta and Mistral join as providers
Three more providers behind the same key, so their models need no separate account or contract.
See it: Xiaomi MiMo, Meta and Mistral join as providers - Site
Site-wide search
Press Cmd K or Ctrl K anywhere to jump to a model, a comparison or pricing.
- Chat
Model council
Ask up to three models at once, then have one more read every answer and tell you where they agree, where they differ and which to trust.
See it: Model council - Site
Sign in with Google, and email codes
Create an account with Google, or confirm your email by typing a code instead of hunting for a link on another device.
- Site
57 model-vs-model comparisons
Price, context and capabilities for pairs of models, generated from the live catalog so the numbers stay current when a provider reprices.
See it: 57 model-vs-model comparisons - API
Bring your own provider keys
Add your OpenAI, Anthropic or other provider key and requests route through it at no extra fee. BYOK traffic shows up in your usage.
See it: Bring your own provider keys - Site
Retired models keep their page
When a provider withdraws a model, its page stays and says what replaced it and how to switch, so an old model id never leads to a dead end.
- Model
- Chat
Replies render as formatted text
Code blocks, lists and tables in a model's answer are shown properly instead of as raw markdown.
See it: Replies render as formatted text - Dashboard
First-run setup for people and for agents
New accounts are walked from sign-up to a first working request, with a separate path if an agent is doing the setup.
- Billing
Credits history
A full record of money in and money out, with labels that say what each line is.
August 2026
7 updates
- Model
xAI and Z.ai, and verified prices for 20 models
Grok and GLM models join, and every listed price was checked against the provider's own rate card, including tiered long-context rates.
See it: xAI and Z.ai, and verified prices for 20 models - Chat
Hosted chat
One prompt, up to three models answering side by side, each answer priced to the fraction of a cent.
See it: Hosted chat - API
companyfabric/auto picks by projected cost
Let the gateway choose the cheapest model that fits each request, and restrict which models it may choose from.
- Dashboard
Usage by date range, to the fraction of a cent
Pick any range, see requests per day, and see spend below one cent as the real figure instead of $0.00.
- Billing
Monthly plans with auto top-up
A flat monthly plan with a usage allowance, and optional automatic top-up so requests keep working when the allowance runs out.
See it: Monthly plans with auto top-up - API
MCP server for agents
Connect Claude Code, Cursor and other MCP clients so an agent can call CompanyFabric models as tools, on your key and balance.
See it: MCP server for agents - Site
CompanyFabric launches
One OpenAI-compatible API for text and code models from every major provider: one key, one balance, exact prices shown before you call.
See it: CompanyFabric launches
Every model above is behind the same key and balance. Free starter credits on signup.
Get API Key