Here are the latest platform updates, upgrades, and improvements from July 2026.
The headline is the new Power plan for heavy users.
The change you will probably feel first is that Home chats behave as in projects. That’s because we upgraded chat in the root of our app and enabled all the advanced capabilities of chats inside the projects.
So you now have the thread memory, documentation upload and AIs have access to project files and the knowledge graph. Expect much smarter AI replies and generally better experience.
Further, the prompt Adjutant opened to every plan, and Super Mind got a stronger synthesis engine and a steadier spine. Here is what shipped.
One thing you may have already read about: nine new models landed in July – GPT-5.6 Sol, Terra, and Luna, Claude Opus 5, Sonnet 5, and Fable 5, Gemini 3.6 Flash and 3.5 Flash Lite, and Grok 4.5. Most took their seats automatically.
The full story is in Nine New Models in Three Weeks.
In the development
True North, part of Suprmind’s AI Anti-hallucinogen system, an automatic dual-layer hallucination prevention
One truth. Everything else is fabrication.
True North moved from the drawing board into a quiet production tuning period. It now runs silently alongside real conversations, in Sequential and Super Mind both, checking high-risk claims as each model generates them – numbers, dates, named entities, citations, financial, legal, and scientific statements.
Two independent layers do the checking, including a verifier outside the model chain, because no AI should grade its own homework. We are tuning it against real production turns before it starts flagging and auto-correcting wrong claims in your threads.
Coming soon
Red Team v2 and the Adjutant are past due design and running daily on a selected number of accounts. Red Team v2 attacks your idea in three turns, across six vectors – financial, technical, reputational, regulatory, operational, and edge cases – then compiles the findings into an exportable risk dossier with a mitigation pass.
The Adjutant is a project-aware strategist that tracks where your project stands, surfaces unfinished threads, and recommends what to ask next. We are fixing what our own runs surface before either reaches you.
Now live
The Power plan
A fourth self-serve plan, above Frontier: $195 a month. Everything in Frontier, plus the highest usage capacity of any self-serve plan, your own API keys (BYOK), unlimited projects, 100 files per project, 15 MB per file, and direct support with same-day response. Built for the people who were topping up every month. Frontier stays the most popular pick for everyone else. See it on the pricing page.
Chat now starts at Home
The root chat – the one you land on – used to be a scratchpad. Now it is backed by a personal project called Home that every account has. That means cross-thread memory, project knowledge, custom instructions, per-AI personalities, and an Overview tab for files, description, and instructions, all without creating a project first. Home is seeded from your onboarding answers, so the AIs know your name, role, and focus from the first message. It stays hidden from your project list and never counts against your project limit. Existing accounts were carried over with full history.
Prompt Adjutant, on every plan
The prompt enhancer that turns quickly typed prompts and questions into structured, professional prompts that produce much better responses from all five AIs. is no longer gated – every plan has it now, trials included.
Super Mind, sharper and steadier
Three upgrades to the mode that runs all five AIs at once. The synthesis honors your response-length setting, so Key Points comes back brief instead of at essay length. And the five models now run on a sturdier transport with more time for the hardest questions.
Your subscription, in your hands
Pause instead of cancel – You can pause your subscription for a month straight from the Account tab, with a reminder email before billing resumes.
Billing at a glance – The Plan tab now shows your card on file, your next billing date, a manage-payment button, and your invoices. Cancel moved to the Account tab, next to Delete Account, with plain copy: canceling keeps full access until the end of the billing period.
A failed payment is not a dead account – If a renewal charge fails, you keep access to the app, your threads, and your billing settings, with a clear banner explaining what happened. Update the card and everything restores automatically.
Usage Boosters, for every plan
Buying a Usage Booster now shows up in your runway immediately, the confirmation says what actually happened, and every plan – Spark included – gets the same one-click top-up when you approach your monthly usage limit.
The Plan tab tells the whole story
In-app plan cards now match the pricing page: complete feature lists, conversation modes, and exactly what each upgrade adds, instead of a few highlights.
Light theme, improved – and it sticks
A full coverage sweep brought onboarding, personalization, the new-project modal, DCI cards, and the sidebar document buttons into light theme.
Smaller improvements
- Recent Chats, expanded – A View more button opens a larger list of recent sessions, readable and click-to-resume.
- Cleaner uploads – One clear allow-list on every plan, with plain rejections instead of silent processing failures. Spark can upload PDFs to project knowledge.
- Enterprise files – The per-project file limit went from 150 to 200.
- Mobile polish – A slimmer collapsed composer, keyboard handling that stops burying your message, and a compact signup questionnaire.
Did you know?
- You can switch modes mid-conversation. Start in Sequential, flip to Debate for a contested call, then drop back. The AIs carry full context across the switch.
- Full Control pins a team. Picking a team by hand applies to one turn on purpose – it stops a heavy team from quietly burning your month. Want it locked for the whole session? Switch on Full Control in chat settings.
- You choose who writes your Master Document. Every template lets you pick the engine, so a legal summary can come from Claude while a data-heavy report comes from GPT.
