nav: icon tabs, and a roomier bid token cap

config: strip whitespace from settings, fixing the prod image search

pipeline: Granite web-search gate, racing the hedge

images on web-search answers, via Tavily

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

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

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

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

qwen: force low bids outside coding

frontier: GPT-5 -> GPT-5.6 Terra

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

web search: freshness-gated OpenRouter web plugin for the agents

bidder answers: go long when detail is asked; raise token caps

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

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

revert verifier: Tencent Hy3 -> GPT-OSS 120B

coding slot: Qwen3 Coder Flash (paid) -> Qwen3 Coder (free-first)

tier-1 reshuffle: V4 Flash (general), Qwen3 Coder Flash (coding), Nemotron (math)

tier-1 lineup: DeepSeek V4 Pro (general), V4 Flash (math), Nemotron 3 Ultra (coding)

swap math bidder: DeepSeek chat -> DeepSeek V4 Flash

swap models: Nemotron 3 Ultra as general bidder, Tencent Hy3 as verifier

deploy hardening: access gate, rate limiting, spend guard, env CORS

eval harness + env-swappable frontier model

hard gate: GPT-5 reserved exclusively for hard STEM/reasoning queries

treat ambiguity as normal, not escalation-worthy

restore 16k frontier token cap; escalations are rare and hard now

remove the soft bid timeout: wait for all bidders

topic toggle takes auction priority when its model bids confidently

hedged speculative draft with user topic toggle

low-confidence bidders skip the reason field; fix truncated deepseek specialty

adaptive frontier token budget alongside effort

adaptive frontier reasoning effort from bid difficulty estimates

calibrate escalation: subjective queries stay tier-1, frontier answers shorter

streaming-first delivery: show drafts immediately, verify in parallel

cut easy-query latency: paid bid endpoints, low-effort verifier, grace-period bidding

speed up inference: speculative drafts in bids, early-exit bidding, shorter retry sleeps

checkpoint: current behavior before latency work

AuctionRouter: cost-aware multi-agent LLM router with auction-based selection and verification-gated escalation