Skip to main content

Feel free to submit feature requests and bug reports :)

NOTE: This is a feedback board for T3 Chat, NOT T3 Code. Please bring T3 Code feedback to Discord, or Github. T3 Code feedback posted to this board will be deleted.

A few quick asks on our part:
1. Search before posting! Your feedback has likely been submitted already
2. If you have multiple features to request, please create separate requests for each.

3. If submitting a bug, PLEASE SELECT “BUG REPORT” WHEN REPORTING

Planned

Search performance

Im experiencing that other providers gives much more precise web search results than T3 Chat. ChatGPT seems to prefer giving web search results when asking about current events, program features, manuals etc. But with T3 Chat I have to give it URLs to product pages, articles etc.

Recommended Tab

Hi, Ive been using t3 chat for quite a while now. The big problem i have is figuring out which model to use for what. i have no idea what the difference between Haiku, Opus, and Fable is much less any of the equivalent from OpenAI, Google etc. I don’t want some dynamic system that tries to figure out what the best model is for the prompt, i just want a recommended tab with a list of the current models and what they are vaguely good at.

Support long decodes for Gemini 4 Argon&Carbon

Title: Support long decodes for Gemini 4 Carbon When using Gemini 4 Argon&Carbon with High effort, I got this error: It looks like the model ran out of tokens during reasoning. Try lowering reasoning, splitting the prompt into smaller tasks, or using a different model. This seems to be caused by Gemini 4's default 65,536-token output limit, which also includes reasoning tokens. However, Gemini 4 supports long decodes of up to 1M tokens (shared between input, reasoning, and output) via continuation tokens. Could T3 Chat support long decodes, perhaps as an optional setting for Gemini 4 models? It would be useful for complex tasks where the model exhausts its reasoning budget before producing an answer.

Planned

Add a "library" of files that the model can look into for extra context

Each profile would have its a limited amount of slots for files that models can retrieve on-demand. For example, I have a profile for job search and often have to upload my CV into every thread when I wanna have an LLM write me cover letters. Would be nicer if the LLMs could simply retrieve my CV from a stash of files that I give it access to.

9

Add support for Haiku 5.5 'xhigh' & 'max', please

Currently, only Instant, Low, Medium, and High are supported for Claude Haiku 5.5 from the dropdown. But benchmarks for Claude Haiku 5.5 on "Xhigh" and "Max" look tasty, "Max" looks better or on par with Gemini 3.8 Flash with "High" (which is the max available option). So switching from Gemini 3.8 Flash to Claude Haiku 5.5 might be an obvious strong money save if all the reasoning options become available (and as long as the context window is <100K, because for context >100K with Haiku 5.5 they bill 5x per i/o token, $0.50/$2.50, which is something you might wanna add to the model info popup btw).

Model Request

BYOK for Fast/Ultra Fast

Would be great to have Fast/Ultrafast as an option with BYOK for OAI/ANT models.

Images or diagrams in results

Other providers more and more gives us images, diagrams and even interactive tools in prompt results. Is this something on the horizon? Also more references in prompts which uses search would be nice.

1
In Progress

Prompt edit

I often edit back my prompt to impove it and get a better answer. To often, my edited prompt isn’t retain and the old prompt is reused like i never modified it.

Model request:GLM 5.3 flash X legal.

z.Ai has released new model named GLM 5.3 flash X legal, it is a multimodel especially trained on legal data, being a law student I am curious to try it. It would be great if we could get it.

Model Request

Metadata / Summary-Based Branching

Problem Branching currently copies raw message history verbatim. In long threads, this: Eats context: Consumes valuable context window space before new work begins. Adds noise: Carries forward superseded code, dead ends, and tangents. Increases latency: Slows down inference with bloated prompt payloads. Proposed Solution Add an option to branch via a compiled context summary rather than copying the raw transcript. The new branch starts lean, seeded with a structured briefing of the discussion up to that point. Workflow Branch selection: When branching, choose between Full Clone or Summary Branch. Auto-synthesis: Generates a brief snapshot of goals, decisions made, active constraints, and current code state. Editable preview: An optional modal lets the user tweak or prune the summary before creating the branch. Clean initiation The summary is injected as the new thread's initial primer, retaining a backlink to the parent thread for traceability. Benefits Maximizes available context headroom for the new branch. Keeps model attention sharp by shedding irrelevant conversational history. Pairs naturally with range-based branching to preserve early context without the raw bloat.

dismiss warning but keep showing next time as default without persistent popup harassment

then when I click X: I want a default option “keep showing”, but allow me to quickly dismiss the banner without being badgered with the supposedly-helpful “are you sure?” pop-up. I need the banner reminder just in case I’ve accidentally left too many settings on, but once I’ve acknowledged it for this chat (by clicking X), I don’t want to be harassed to click “keep showing” it every single time.

2
Gathering Interest

Auto-updating chat title

If a current chat goes off topic or the user moves to a different topic, the title of that chat stays the same with whatever the user first sent in that chat. I want an auto-updating chat title that updates every X amount of messages (either that can be configured by the devs or by the users) so we know the chat title isn’t outdated. [and yes i know there is a regenerate chat title option, but i want it to be auto]

Chat UX
Gathering Interest

Auto-Handoff

A simple button to automatically have the current model write a handoff of the chat, create a new thread and send the handoff in that chat. Helpful when a thread is getting long but your problem hasn’t been solved.

Chat UX
Gathering Interest

Auto Swap Models

It would be wonderful if the moment I enter burst overage, I could customize it so that it would automatically (or ask to swap) to a model of my choosing.

1
Planned

Increase Contrast between states in mobile model filter buttons

The contrast between selected and not selected of the model selector filter buttons is too low, making it harder to determine if a filter was selected or not

Mobile & Desktop

Profile switching on mobile

Whenever switching profiles on mobile, you currently have to go to the sidebar, click your profile and then switch the profile. Since there is no way to set another default profile or remember the last used profile, I reckon it would be a good addition to add it somewhere in the header. Maybe next to the chat/canvas switcher or within the chat settings ui.

4Chat UXMobile
Gathering Interest

Custom Model Presets

TL;DR Optimizing Model capabilites is sluggish. Add a “Create Model Preset” Option to model selection. Allow to save “model + reasoning + number of searches“ as a named preset. Problem Trying to optimize for Token efficiency, it is sluggish to adjust Model capabilities based on my upcoming Prompt. I find myself configuring reasoning and search all the time. I also find myself often times checking those settings when switching to another model before I write my prompt. This adds friction. Goal I would like to have the ability to configure Model presets. e.g. a research preset, that would possibly be something cheap/cheaper with maximum searches enabled. e.g. “deepseek v4 flash, low reasoning, 5 searches“. Other presets could be simply high reasoning and include something like “kimi k3, high reasoning, no searches“ or any in-between. Maybe “creative support & planning“, which could be “kimi k3, low reasoning, 2 searches“. These are presets, that I can just think of right now. What I find sluggish is, that when I made a research request with a cheap model, it has the full search capabilites. Maybe I now want a more intelligent model, but don’t strip search completely from it, if it decides, that it has to look up information. Switching back and forth requires a lot of clicking in the ui. It is cumbersome when I want the same model, but with different estimated token cost, generally adjusted for the task. What would I use it for Planning Trips or researching would allow me to quickly select my preconfigured favorite models for the Task at hand. I could quickly prompt a detailed search based on some criteria, e.g. looking for interesting spots on the road when I plan a roadtrip from point A to point B. Right afterwards use the same model to rank or filter those models based on feedback and possibly reason over a clever schedule. Maybe still allowing for some search to lookup current conditions, timed events, known road blocks etc. Or simply ditch the research and try to develop a detailed goals & wishes, dos & don’ts list. Same pattern can be applied to learning new skills, researching new topics etc.

Chat UXAPI & IntegrationsAccessibility
Gathering Interest

Model request: GPT 5.6 Luna xhigh or max reasoning levels

It would be great to add support for GPT‑5.6 Luna xhigh or max reasoning levels in T3 Chat. OpenAI recently changed the pricing for its Luna and Terra models, making GPT‑5.6 Luna xhigh and max reasoning levels considerably cheaper than many other available options. Adding these models could reduce costs while maintaining a high level of intelligence.

1Model Request