health: report whether image search actually works; don't eat $ as math

citations: space before the link, and a hover preview card

images: fixed height, natural width — stop cropping entirely

images: rounded uncropped tiles, and a fallback search query

ui: darken the composer to match ChatGPT's contrast, not its hex

ui: icon copy buttons, orange confirmation, ChatGPT composer grey

pipeline: give the hedge the web, instead of skipping it

images: search the answer, not the question; larger uncropped tiles

chat: stop, regenerate, inline prompt editing, copy answer

images on web-search answers, via Tavily

rich teaching answers, math rendering fix, emoji hard cap

chat tips: friendlier copy with GAVL as the actor

web-need detection at t=0, shared answer formatting, friendly copy

answer formatting + UI: rich-answer format prompt, gpt-oss labels, input polish

router: GPT-OSS 120B generalist + Llama 4 Scout verifier; greeting fix; input UI

user message bubbles: rounded-full to match the input pill

input typed text stone-200 -> stone-100 (closer to white)

toggles up (mt-3->mt-1), bigger gap to input box (mt-2->mt-4)

nudge topic toggles right (pl-4)

remove 'topic' label next to the toggle buttons

input box: rounded-3xl -> rounded-full (pill), more side padding

brighten input placeholder text stone-600 -> stone-400

remove borders from topic toggles and chat/metrics tabs

remove input box border; subtle bg fill (rounded, borderless)

richer /explain, new copy: splash tagline, greeting, bot animation

tips as rotating input placeholder; boss opener holds longer; pixel tagline

chat empty state: drop NEW GAME + box, keep only the (bigger) tagline

tips linger ~10s into a query instead of vanishing at first token

rebrand to GAVL

loading tips + /explain command

narrow answer bubbles further 68% -> 58%

answer body text stone-300 -> stone-200 (closer to white)

answer links open in a new tab (target=_blank, rel=noopener)

web search: GPT-style rotating ticker instead of status-line-only

web search: tell verifier answer is web-grounded; add 'searching' stage

move copy button from whole-message to per-code-block (top-right)

input box: rounder corners (rounded-xl -> rounded-3xl)

bid_timeout 20->30s; narrow AI bubbles to 68%; round the input box

answer headings: white bold instead of colored (orange)

tier-1 lineup: Gemma 4 (general), Qwen3 Coder (coding), DeepSeek V4 Flash (logic/math)

fix math rendering: strict LaTeX delimiters in frontier prompt + forgiving KaTeX

tier-1: DeepSeek V4 Flash (general), Qwen (coding), Skyfall 36B V2 (creative)

general slot: Llama 3.1 70B -> Llama 4 Maverick

general slot: Hermes 4 70B -> Llama 3.1 70B Instruct

fix literal 'Subheading' echo in prompts; bold back to white

general slot: Gemini 2.5 Flash Lite -> Hermes 4 70B

style models' bold labels as sub-headings (larger + orange) with wider point gaps

answer typography: larger sub-headings with gap above; drop inline bold

narrow assistant/live bubbles to 80% width

tier-1: Gemini 2.5 Flash Lite (general), DeepSeek V4 Flash (math), Qwen3 Coder (coding)