Two months of work. Three new frontier models added as soon as they launched, AI Teams finally deployed, app and website translated into three new languages, a Document Intelligence pipeline rebuilt, a bigger context window, and one quiet win: Every conversation now costs less to run.
This change log headliner is AI Teams. You can now choose which models make up your lineup and which team answers each prompt in each turn, with a single click. Do not want to manage it? The smart Auto selector reads each prompt and picks the right team for you.
Everything else is below.
Now live
– AI Teams
Pick the models that run your conversations. AI Teams groups the five providers into three teams, and you choose which model from each provider fills each slot. Set them up on the AI Teams settings page.
- The A-Team – deep deliberation for strategic decisions, novel problems, and high-stakes calls.
- Operators – the workhorse team for real tasks that do not need maximum brainpower at maximum cost.
- Daily Drivers – capable on their own, these models compound into a fast team that handles 70 to 80 percent of everyday work: parsing, extraction, and data tasks.
Out of the box, the A-Team runs Claude Opus 4.8, GPT-5.4, Gemini 3.1 Pro, Grok 4.3, and Sonar Reasoning Pro. The defaults are not guesswork. We built them on our own accounts and tuned them against a huge amount of real work on the platform. Customize any slot you want.
Each team has its own reasoning depth set automatically, and the Deep Thinking toggle forces full reasoning whenever you need it.
Choose which team replies per turn from the chat settings panel. Leave it on Auto and the selector reads each message – task, complexity, length, attachments – and routes it to the best-fit team. Pick a team by hand and it applies to that turn, then hands back to Auto. To keep one team locked across the whole conversation, switch on Full control.
Available on Pro, Frontier, and Enterprise.
One note: GPT-5.5 is available on Frontier plans, but it is not selected as the default model in the AI Team. Its cost sits on par with Claude Opus 4.8, and running both at once burns through usage fast. Add it yourself if you want it. You have full control.
– Lower cost on every conversation
We shipped 37 improvements to prompt caching and to how images and documents are handled, much of it focused on Claude. The cost of each conversation turn dropped, with no change to the models or what they can do.
– Now in German, French, and Spanish
Suprmind works end to end in three more languages. The app interface, onboarding, settings, the homepage, and the help content are all localized for German, French, and Spanish. The AIs already replied in your language. Now the whole product does too. Switch languages from the menu.
– Document Intelligence V2
We rebuilt how Suprmind reads your files.
Drop a large, messy document into the chat and the pipeline cleans it, parses it, converts it to markdown, and annotates pages, subheadings, and chapters. A model with a two million token window then reads the whole thing and writes a Doc Intel Brief. The document is vectorized and available to all five AIs for deep analysis.
Each file shows its status as it moves through the pipeline – Indexing, Ready, or Failed – so you know when it is ready to use. You can search your files by name in the Files panel. And more of your documents get the full treatment now, because we lowered the size threshold for deep processing.
– The latest models, and a bigger context window
We keep the lineup current. Every provider is on its newest models – Claude Opus 4.8, GPT-5.4, Gemini 3.1 Pro, Grok 4.3, and Sonar Reasoning Pro. GPT-5.5 is now available on Frontier and Enterprise.
Claude also runs on a much larger context window, one million tokens, without a previous double cost penalty for context over 200.000 tokens. Very long threads and large documents stay fully in context, so nothing important falls off the back of the conversation.
– Also new and improved
- Larger file uploads. Pro and Frontier both got higher upload limits.
- Cleaner composer. Paste a large block of text and Suprmind turns it into a .txt attachment automatically, so the message box stays readable.
- Live search for Gemini. On Pro and up, Gemini now grounds its answers in current Google Search results, alongside the fresh-data lookups from Perplexity and Grok.
- In-app announcements. Product updates and notices now show up inside Suprmind, not only by email.
- Screenshots in support. In-app support can now read images you send, so you can show a problem instead of describing it.
Coming soon
– The Adjutant
A project-aware strategist with you on every project. The Adjutant watches the work in your project and surfaces the right context, prompts, and next steps before you go looking for them. It is in validation with Enterprise users now, and a wider rollout is coming.
– MCP Connectors
Connect your tools, and let the AIs use them. We are finishing testing on connectors for the Google suite, Notion, Google Analytics, and many more through Zapier. Once a tool is connected, all five AIs can read from it and act through it inside the same conversation. This is real agent orchestration. The team does not just discuss your work, it can go and do parts of it.
– Red Team v2
The biggest upgrade to any mode yet. Red Team v2 runs a full round of attacks on your idea, then a defense turn where the case gets argued back, then a sharpened second round that goes after whatever survived. In the 4th turn, the user gets the synthesis and the Adjuicator’s Risk Posture Dosier. You get a much harder stress test and a clearer read on which risks actually hold up.
Platform maintenance
Maintenance, fixes, and steady polishing run in the background every day. We hope you are feeling the difference across the app.
Did you know?
Two ways to get sharper answers and stretch your usage:
- You do not need all five models on every turn. Click the provider pills above the chat box to include or drop any AI, or tag one directly with @ to send a question to it alone.
- Long threads cost more per turn, because each turn carries the weight of everything before it. When your direction shifts, start a fresh session instead of pushing one thread to its limit. The AIs focus fully on the new goal, and the answers get better. For the record: one session this June reached 107 turns, up from the old high of 57. You can run that long. You rarely need to.
