Changelog

The assistant now verifies answers by reading official pages directly. When asking about service status (like Royal Mail or Canada Post), flight updates, or government services, it fetches the actual page to give you accurate, real-time information instead of guessing.

Upgraded web search to use Perplexity by default for better quality results with clear source citations. Perfect for research, news, and finding authoritative information. The assistant automatically uses the right search tool based on what you need.

Just share a URL and ask the assistant to read it. Gets clean content without ads or clutter, perfect for analyzing articles, documentation, or research papers.

Updated to the latest AI technology stack with better performance and compatibility. Navigation between pages is now smoother and more responsive.

Fixed several technical issues to make everything run more smoothly and reliably.

Added Anthropic's latest Claude Sonnet 4.5 with superior coding, reasoning, and instruction following. Now featured at the top of the model picker with extended thinking mode and 200K context window.

Added DeepSeek's latest experimental model with DeepSeek Sparse Attention (DSA) for improved long-context efficiency. Features 163K context window with 65K output at 50%+ lower cost.

Complexity-aware model selection automatically picks the best model for your task based on benchmark data from over 280 AI models.

See tokens per second (TPS) for every response across all providers, helping you understand model speed and efficiency.

Added latest September 2025 models from OpenRouter including updated reasoning and coding specialists.

Improved authentication telemetry for faster troubleshooting and more reliable sign-ins.

Updated social media preview images and artwork for better sharing across platforms.

TPS timer now starts on first data chunk and estimates client-side for instant feedback without page refresh.

Added xAI's Grok 4 Fast with its 2M token context window and optional reasoning traces. It is now selectable in the model picker and Auto Mode routes to it for tougher reasoning tasks.

Promoted Grok 4 Fast across featured lists and fallbacks while keeping Grok 4 (0709) as the heavyweight fallback when you need deeper analysis.

Now featuring Qwen3 Next (instruct & thinking), Qwen Plus 0728, Qwen3 30B & 235B Thinking, StepFun Step3, Moonshot Kimi 0905, Sonoma Alpha previews, Seed OSS 36B, Cogito V2 previews, and NVIDIA Nemotron Nano 9B v2.

Each new model highlights its strengths (reasoning, coding, multimodal, long-context) so you can pick the right fit faster.

Every addition was live-tested so you can jump in without worrying about availability surprises.

Add “forgot password” and email verification so you can securely get back into your account.

More reliable guest and OAuth sign‑ins with clearer messages if something goes wrong.

Cleaner plan switching and smarter currency handling for a faster checkout.

Refreshed pages (Terms, Privacy, Changelog) and general UI copy/visual updates.

Resolved occasional issues with email handling and sign‑in on certain devices.

One‑click sign‑in options, plus classic email accounts. Stay as a guest if you prefer.

Ask questions that need fresh info. We search the web and summarize with sources.

Attach screenshots, photos, and PDFs so the assistant can see and discuss them.

Auto Mode chooses a great model for your task and enables “thinking” when helpful.

Hundreds of models supported across providers, with better defaults for speed and quality.

Access to the newest releases (like GPT‑5, Claude 4.1, and Grok 4) across providers.

Cleaner interface and easier flows across devices for a nicer feel.

Updated model comparison pages so you can pick with confidence.

General performance and reliability improvements across the app.

Let the AI take a more careful, step‑by‑step approach on harder tasks.

A guided mode that pulls from multiple sources and returns clear, cited summaries.

Lots of reliability fixes so sessions run smoothly without interruptions.

Quicker first responses thanks to routing and caching improvements.

Access Anthropic's latest Claude with stronger reasoning and coding skills.

New open‑weight GPT OSS options for creative work and problem‑solving.

The router now picks improved defaults so you get great results with less effort.

More choices for reasoning and coding tasks with excellent language coverage.

Improved formatting and highlighting makes generated code easier to read.

Better handling for images and documents so you can keep working without hiccups.

Resolved rare cases where messages didn’t send correctly during streaming.

Uploaded files stay “in mind” longer so follow‑ups feel more natural.

Smoother handling of bigger files with better caching.

Fewer interruptions while the AI is replying.

Start a chat, add images and PDFs, and get helpful answers.

Stop an answer mid‑way when you already have what you need.

See when tools run and how results show up in the conversation.

A round of fixes for uploads, scrolling, and login edge cases.