Every door we've opened lately

DeepSeek V4.1 Flash moves to the 1017 checkpoint

The existing DeepSeek choice now uses DeepSeek V4.1 Flash 1017. Saved chats and assistants follow automatically, with the same tools, image input and 32,768-token output limit.

• The standard route stays pinned to Fireworks through AI Gateway with zero data retention and no provider fallback.
• Standard pricing is $0.30 input, $1.20 output and $0.006 cached input per million tokens. The optional GreenPT route uses the same exact checkpoint at €0.36 input, €1.45 output and €0.04 cached input per million tokens.

Keep the answer you prefer after comparing models

Choosing the second answer now saves it before closing the comparison. If saving fails, the answer stays open so you can retry without generating it again. You can compare the saved answer again later.

Sign in without leaving Create Account

Already have an account? Entering your existing email and correct password on Create Account signs you in directly. If those details do not work, stay on the same screen to try again, reset your password or continue with Google.

Close search easily on your phone

Search chats now has a larger close button that stays clear of the status bar, so you can return to your sidebar without choosing a search result.

Argon shows its image-generation capability consistently

Gemini 4 Argon now shows its image-generation capability consistently in model choices and API capabilities. Image creation uses Magicdoor’s image tools, not native Argon image output.

Keep useful matches when chat search is incomplete

Chat search keeps the matches it has already found when another part of the search fails, and tells you when some chats could not be searched. The assistant also avoids repeatedly retrying a failed search in the same reply.

Clearer privacy wording for shared-content reports

The privacy policy clarifies that a shared-content report includes a link to the reported shared conversation.

A cleaner image library

The image library now matches the rest of Magicdoor in light and dark mode. Phones show two full-width previews per row, search and filters sit in one compact bar, and New image is always at the top. Tap any image to see its full prompt, edit it or reuse the prompt, and Animate stays on every image.

Images opens your library, and settings are one tap apart

Images in the sidebar now opens your image library, with a New image button whenever you want to make another. The video library has a matching New video button. Account, Usage and API keys now share tabs, so you can move between them without going back to chat.

Clearer chats, with recorded reply costs

Each answer now names the model that wrote it and shows its recorded cost when available, including images created in the chat. The chat header shows the running cost of the chat; open it to see the cost by model, how full the context is, and the option to compact long chats. Cost details stay scrollable on smaller screens. Your balance stays visible in the sidebar, and image generation shows the price on the Conjure button.

A refreshed look across the website and sign-in

The homepage, pricing, comparisons, guides, sign-in and account pages share a cleaner design that follows your light or dark setting. The homepage now shows image and video creation next to chat, with a sample image made by each image model.

Meet Gemini 4 Argon

Choose Gemini 4 Argon for text, images, PDFs, search and Canvas with low, medium or high reasoning effort.

• 1M-token shared context with up to 32,768 output tokens.
• Introductory token rates through November 30, 2026.

Keep follow-ups during chat recovery

A follow-up sent while a recovered answer finishes saving stays in the same chat instead of silently disappearing. Stop still cancels a queued follow-up.

Model and pricing guides reflect current choices

Compare the current chat models, including Claude Haiku 5.5 and Mistral Large 4, with clearer usage prices. Retired Gemini Flash and MiniMax models are no longer recommended as current choices, and Basic budget examples distinguish the $2 first-month balance from $1 on renewal.

Claude Haiku 5.5 joins Chat

Claude Haiku 5.5 replaces Gemini 3.7 Flash and MiniMax M3 in the picker. Existing chat history stays readable; choose a current model to send a new request. Haiku starts at $0.10 input and $0.50 output per million tokens for prompts up to 100,000 tokens, with higher rates for longer prompts.

Mistral Large 4 (Le Chonk) joins Chat

Choose Mistral Large 4 (Le Chonk) for text and image conversations. It is a public preview, supports tools and offers a 524K-token context window. It is an additional choice; your existing model selection remains available.

Clearer image and video failure explanations

When an image or video model rejects a generation through its safety filter, Magicdoor now explains the reason and identifies the model provider. Failed videos keep that explanation when you return to them.

GPT-6.1 Sol replaces GPT-6 Sol

GPT-6.1 Sol is the balanced OpenAI option for complex work in Chat and Auto. Existing GPT-6 Sol chats continue on 6.1. Standard input and output rates stay at $2 and $10 per million tokens; cached input falls to $0.10 per million.

Claude Sonnet 5.5 is here

Choose Claude Sonnet 5.5 in chat for writing, coding, research, and document work. It replaces Sonnet 5 at the same listed usage rates, and existing Sonnet chats continue on the new model.

Code-generated files download without broken previews

Files created while Claude or GPT runs code remain downloadable in the conversation without appearing as broken images in the library or sidebar.

Send feedback inside Magicdoor

Send ideas and bug reports from your account menu, or report a shared conversation from its header. You choose what to include; private conversations and attachments are not added to your message.

Interrupted chat replies are easier to recover

If a reply stops early, Chat now clears the loading state, keeps any answer text already shown, and offers Retry instead of leaving an empty response or a stuck indicator.

Claude can work with your data in Chat

Ask Claude to run calculations, analyze CSV or Excel files, and create downloadable files in the same conversation. Text and code files also work without switching modes. Running code or sending a data file adds about $0.0042 per Claude turn, plus token usage.

Grok 4.7 is here

Choose Grok 4.7 in chat for text and image questions with a 500K-token context window. It replaces Grok 4.6 at the same listed usage rates, and existing Grok chats continue on the new model.

Claude Opus 5.5 and GPT-6 Sol and Luna arrive

Claude Opus 5.5, GPT-6 Sol, and GPT-6 Luna replace their direct predecessors in the same picker positions. Existing chats, assistants, links, preferences, and API integrations continue on the new models automatically.

• GPT-6 Sol is built for complex coding and agentic work; GPT-6 Luna is the low-cost choice for focused, high-volume work
• Claude Opus 5.5 brings a 1M-token context window and provider-managed adaptive thinking for long-running coding and knowledge work
• Auto now uses Claude Opus 5.5 for judgment-heavy work and challenged answers, while GPT-6 Luna keeps code execution and file workflows compatible

Smarter large PDF uploads

Chat now safely optimizes large PDFs on your device before upload. If a document or combined chat request is still too large, you get clear guidance and your message and attachments stay ready to edit and retry.

Start for $6, or choose more monthly usage with Plus

Basic now starts at $6 with $2 usage balance in your first month, then $1 each month. The new $12 Plus plan includes $7 usage balance every month. Both plans give you the same models and features, with no extra first-payment charge.

• Included balance resets each month. Purchased top-up balance never expires.

Smoother scrolling in mobile chats

Scroll back through earlier messages while an answer is being written without getting pulled to the bottom. Return to the bottom to follow the answer again.

Pick up an interrupted reply

If your connection drops while an answer is being written, chat can catch up and continue without sending your message again. Returning to a finished reply restores the answer, while ordinary conversations keep streaming immediately.

Create a Canvas just by asking

Ask a Canvas-capable model to create a document and it opens ready to use. On mobile, Back to chat keeps the Canvas attached, so your next message can keep reading and editing it until you switch or reset the chat.

GPT Image powers your default image workflow

New image sessions start with GPT Image 2.5 Sunburst. Images created or edited in chat use GPT Image 2.5 Flare at Low quality by default, with other aspect ratios and higher quality available on request.

Web searches now use Exa exclusively

Your selected chat model uses Exa to find sources and writes the answer. Perplexity is no longer available for new searches; previous Perplexity answers and citations remain visible in your chat history.

More reliable replies in Auto

Auto can try another service or model when the current one is slow or unavailable, helping your answer finish without starting over yourself. Problems with an individual tool stay beside that tool so the conversation can continue.

• A quiet “Trying again…” status appears only when Auto recovery takes more than a moment
• Your answer still shows the model that responded, and completed tool work is kept during recovery
• If recovery cannot finish, your partial answer stays visible with a Retry action

Project chats keep their generated titles

New conversations inside projects now replace the temporary “New Chat” label reliably, even when generating the title takes a little longer.

DeepSeek upgrades to V4.1 Flash

DeepSeek V4.1 Flash replaces the 0731 release with a 1,048,576-token context window and zero-data-retention routing. Existing DeepSeek V4 Flash chats continue automatically.

• Costs $0.22 per 1M input tokens, $0.007 per 1M cached input tokens, and $0.66 per 1M output tokens on the pinned Fireworks route
• Supports vision, Magicdoor tools, web search, image generation and editing, Canvas, memory, and Oggie with a 32,768-token output cap
• Pins every request to Fireworks with ZDR required and fails closed if that route is unavailable

GLM-5.3 models join renewable inference

GLM-5.3 and GLM-5.3 Flash now have the Renewable option, so you can choose lower-impact inference for both Z.ai models alongside Kimi K3.

• Adds Renewable to GLM-5.3 and GLM-5.3 Flash in the model picker
• Keeps image attachments available for GLM-5.3 Flash and Kimi K3 on the renewable route

DeepSeek V4.1 Flash joins renewable inference

DeepSeek V4.1 Flash now has the Renewable option alongside GLM-5.3, GLM-5.3 Flash, and Kimi K3, with its own GreenPT pricing and image input support.

• Adds Renewable to DeepSeek V4.1 Flash in the model picker
• Keeps image attachments available on both the Fireworks and GreenPT routes

Meet Auto: let Magicdoor choose the model

Spend less time choosing a model and more time getting things done. Auto is an experimental chat mode that chooses a model for each message and can bring in a stronger model when a request needs more reasoning.

• Choose Auto in the chat mode menu, or open Pick a model whenever you want to choose for yourself
• See which model answered every message, and use Compare to replay a response with another model
• Make Auto your default for new chats in Customize Magicdoor
• Auto is experimental. Try it with your everyday tasks and let us know where it gets the choice right or wrong

A calmer view of model reasoning

Reasoning now stays collapsed while an answer is generated, with a live timer and a final thought duration. Open it whenever you want without it snapping shut when the response finishes.

Easier canvas controls on your phone

Canvas has one easy-to-tap close button, with controls kept clear of the iPhone status bar and home indicator. Copy, undo, redo, and view controls stay visible on touch screens.

GPT Image 2.5 adds Flare and Sunburst to Images

Choose Flare for faster everyday generation and editing, or Sunburst for precision-focused image work. Older GPT Image selections continue automatically on Flare.

• Both models offer Low ($0.012), Medium ($0.047), High ($0.128), X-high ($0.25), and Max ($0.50) quality tiers; Flare defaults to Medium and Sunburst defaults to High
• Both support the existing upload, paste, and prompt-first editing flow with up to four outputs
• Saved GPT Image 2, 1.5, and 1 workflows continue on GPT Image 2.5 Flare

Preview HTML in Canvas

Keep writing formatted documents in Canvas, or use Code and Preview for a webpage. HTML code blocks stay part of your document until you choose to preview them. Refine a page in chat and see it update when the assistant finishes writing. Preview supports styled pages and CSS animations; scripts and external content are not supported.

Canvas can read your document, edit a passage, and write live

Your assistant can read the full canvas and edit individual passages without rewriting the rest. Watch new documents appear as it writes. Conflicting edits are stopped to preserve your changes, and saving is more reliable when you clear a document or switch chats.

A cleaner launch and full-screen backgrounds on iPhone

The next iPhone app update adds a Magicdoor launch screen and extends backgrounds to the screen edges. Menu and chat controls stay clear of the status bar and home indicator in portrait and landscape.

• The chat background continues behind the status bar and home indicator without contrasting white strips.

Steadier scrolling in mobile chats

Chat controls stay visible when the keyboard opens. Follow new replies at the bottom, or scroll back to read earlier messages without being pulled away as a response arrives. Sending a message brings you back to the latest reply.

• On mobile web, chat, image and video controls fit the visible screen when the keyboard opens.
• Sending a message on your phone keeps the keyboard closed while the reply arrives.

Recraft upgrades to V4.1 in Images

Recraft V4.1 replaces Recraft V4 for art-directed image generation, while the Images picker is simplified by retiring Flux.1 Kontext Pro. Older image workflows continue automatically on current models.

• Adds official recraft-ai/recraft-v4.1 generation at $0.04 per image
• Keeps the existing upload, paste, and prompt-first editing flow unchanged
• Routes saved Recraft V4 and Flux.1 Kontext Pro workflows to supported replacements

GPT-6 Astra opens a new premium door

GPT-6 Astra is now available to subscribers for OpenAI's hardest end-to-end work. GPT-5.6 Sol remains available as the default and cheaper OpenAI option.

• Adds a 1.05M-token context window with up to 922K input tokens and 128K output tokens
• Supports vision, web search, image generation, Canvas, reasoning, code interpreter, and PDFs
• Uses standard rates of $10 input, $1 cached input, $12.50 cache write, and $50 output per 1M tokens

A cleaner home for project chats

Expand projects in the sidebar to open their chats, or use the new-chat button to start a conversation. Project pages now have a simpler chat list and compact instructions with Project settings beside the title. Recently opened projects reopen without waiting, and longer chat lists load as you browse.

• Rename, pin, or delete a chat directly from its menu inside an expanded project.

GLM-5.3 and GLM-5.3 Flash arrive in Chat

GLM-5.3 replaces GLM-5.2 for long-horizon coding and reasoning, while the new low-cost GLM-5.3 Flash adds image input. Older GLM chats continue automatically.

Pricing and company details are easier to find

Magicdoor now has dedicated Pricing and About pages, plus refreshed comparisons, so costs, included models, and company information are easier to find.

Claude Fable upgraded to 5.1

Claude Fable 5.1 replaces Claude Fable 5 across magicdoor.ai. Saved chats using Fable 5 continue on the new model automatically.

• Keeps the 1M-token context window, 128K-token output cap, and subscriber access
• Reduces cached-input pricing from $1.00 to $0.25 per 1M tokens

Clearer upgrade prompt when credit runs out

Chat and image actions now show a focused upgrade prompt when your balance hits zero, with your draft prompt preserved.

• Chat drafts and image prompts are restored after a failed send instead of being lost
• Active subscribers with an exhausted balance are routed to Usage to top up, not to subscription copy
• Upgrade links use the server-provided destination so checkout stays consistent

A clearer Subscribe page

The Subscribe page now shows plainly what the base plan includes: 15 chat models, 8 image models, $6/month with $1 credit included.

• One subscription for 15 chat models and 8 image models instead of separate provider plans
• $6/month with $1 usage credit included; top-ups only when you need them
• No rate limits or cooldowns

GPT-5.6 Sol drops up to 33% in price

Following OpenAI's promotional price cut, GPT-5.6 Sol now costs $4 per 1M input tokens and $20 per 1M output tokens on Magicdoor (previously $5/$30). Promo pricing holds through at least November 21, 2026.

• Input tokens drop from $5 to $4 per 1M (-20%), output from $30 to $20 (-33%)
• Cached input falls to $0.40 per 1M tokens
• Applies automatically in Chat - no action needed

Qwen3.8 Max replaces DeepSeek V4 Pro in Chat

DeepSeek V4 Pro is being deprecated by our provider, so the chat lineup now includes Alibaba's Qwen3.8 Max — a vision-capable open-weight flagship for coding and agentic work. Existing DeepSeek V4 Pro chats move over automatically.

• Adds Qwen/Qwen3.8-2.4T-A95B with image input, search, reasoning, and a 256K context window
• Routes saved deepseek-ai/DeepSeek-V4-Pro chats to Qwen3.8 Max so conversations keep working
• Prices Qwen3.8 Max at $2.50 per 1M input tokens and $6.25 per 1M output tokens

Renewable-powered AI is now an option

GLM-5.2, Kimi K3, and DeepSeek V4 Flash can now run on 100% renewable-powered infrastructure in the EU with one simple toggle.

• Shows a Renewable toggle whenever the selected model supports the lower-impact route
• Adds a setting to prefer renewable inference automatically while keeping other models and Magicdoor tools unchanged
• Puts per-response energy estimates into context with familiar AI, LED-bulb, and wash-cycle comparisons

Model lineup and pricing refreshed

The chat lineup now includes the latest selected models, while provider pricing has been refreshed so usage costs stay accurate — including GPT-5.6 Luna at its new lower rate.

• Adds GPT-5.6 Luna and Sol, Claude Sonnet 5, Claude Fable 5, Claude Opus 5, DeepSeek V4 Pro, DeepSeek Flash, Gemini 3.7 Flash, Grok 4.6, Kimi K3, GLM-5.2, and MiniMax M3 to the current lineup
• Retires Gemini 3.1 Pro from the model picker while keeping older chats readable
• Updates token and image-generation pricing across OpenAI, Together AI, Google, xAI, and Replicate models
• Moves internal jobs to GPT-5.6 Luna for cost-efficient automated work

Chat opens faster

Magicdoor now loads your chat preferences alongside the subscription check, removing an extra browser round trip before chat becomes usable.

• Opens new and existing chats without a separate preferences loading step
• Keeps your saved default model and chat settings ready on first render

Seedream upgraded to 5 Pro

Seedream 5 Pro replaces Seedream 4.5 in the Images workspace with generation and editing from up to 10 reference images.

• Adds official bytedance/seedream-5-pro generation and editing
• Supports up to 10 reference images through the existing upload and paste flow
• Uses affordable 1K output at $0.05 per image

Chat tools can complete longer workflows

Magicdoor chat now gives multi-step tool workflows up to ten model-generation steps, so longer research and action chains have more room to finish.

• Applies the higher default across OpenAI, Anthropic, Gemini, Grok, and Together models
• Only uses the additional steps when a tool workflow needs to continue
• Keeps parallel tool calls available within each model-generation step

Claude Opus upgraded to 5

Claude Opus 5 is now available on magicdoor.ai as Anthropic's premium everyday model, with Opus pricing unchanged.

• Routes new Opus chats to claude-opus-5 while keeping older Opus chats readable
• Keeps Opus pricing at $5 input and $25 output per 1M tokens, with the same cache pricing
• Preserves subscriber access, model switching, files, vision, and tool workflows

Web search now uses Exa retrieval

Magicdoor chat can now retrieve fresher, more relevant web sources with Exa while the model you selected writes the answer and links its sources.

• Uses semantic web retrieval for current and source-dependent questions
• Keeps the selected chat model responsible for the final answer
• Shows numbered source links in chat and keeps Perplexity available as a fallback

Chat replies start faster

Chat responses now begin sooner by reducing the setup time before streaming starts.

Grok upgraded to Grok 4.5

Grok 4.5 is now available in magicdoor.ai with xAI reasoning, image input, a 500K-token context window, and usage pricing from $2/$6 per 1M tokens.

• Routes new Grok chats directly to xAI's grok-4.5 model
• Keeps saved Grok 4.3 and grok-4-latest values resolving to Grok 4.5
• Applies $4/$12 long-context pricing at 200K prompt tokens and above
• Moves the canonical model page to /models/xai/grok-4-5 while preserving the old Grok 4.3 URL

Existing chats open faster

Opening an existing chat now reuses its title from the messages request, removing an extra database lookup so conversations finish loading sooner.

Perplexity web research streams reliably

Perplexity research invoked from other chat models now handles citation-bearing stream chunks reliably and reports interrupted searches instead of leaving an apparently complete answer cut off mid-sentence.

GPT-5.6 Sol and Luna replace GPT-5.5 and Mini

GPT-5.6 Sol is the new OpenAI flagship and GPT-5.6 Luna is the efficient OpenAI option in magicdoor.ai. Legacy GPT-5.5 and GPT-5.4 Mini chats resolve to the new models automatically.

• GPT-5.6 Sol replaces GPT-5.5 as the first default model in the picker
• GPT-5.6 Luna replaces GPT-5.4 Mini as the efficient OpenAI option
• Keeps legacy GPT-5.5 and GPT-5.4 Mini chats resolving to the current models
• Updates model pages, recommender, marketing copy, and onboarding to show GPT-5.6 Sol and Luna

Saved chats start streaming faster

Saved chats can start streaming sooner because magicdoor.ai now skips a redundant ownership read before the response begins.

Claude Fable 5 is available again

Claude Fable 5 is back in magicdoor.ai for subscribers, with Anthropic routing, a 1M-token context window, and an explicit 128K-token output cap for long responses.

• Restores Claude Fable 5 to the chat model picker and integrations API
• Keeps Claude Fable 5 subscriber-only
• Refreshes model pages, public model counts, llms.txt, and onboarding copy to show 15 chat models

New chats start faster

New chat threads now begin streaming before the saved conversation finishes setting up, so first replies can start sooner while draft chats resume more reliably once they are persisted.

Perplexity direct chat streams reliably

Direct Perplexity model chats now handle Perplexity streaming chunks with citations more reliably, preventing a stream validation crash before answers reach the chat.

Claude Sonnet upgraded to 5

We updated magicdoor.ai to use Claude Sonnet 5 with the claude-sonnet-5 API model ID, 1M-token context window, and standard Sonnet pricing.

• Routes new Sonnet chats to claude-sonnet-5
• Keeps legacy claude-sonnet-4-6 saved values and model-page links resolving to Claude Sonnet 5
• Refreshes model pages, resource guides, marketing copy, recommendations, and API examples to show Claude Sonnet 5

GLM-5.2 replaces GLM-5.1 on Together AI

GLM-5.2 now replaces GLM-5.1 in magicdoor.ai with Together AI routing, a 262K-token serverless context window, 131K-token output cap, and verified usage pricing.

• Uses zai-org/GLM-5.2 as the selectable Z.ai Together AI chat model
• Routes saved GLM-5.1 values to GLM-5.2 for compatibility
• Includes verified input, cached-input, and output pricing for cost tracking

Chat history updates more reliably

The sidepanel now refreshes from the latest saved chat history when you create or finish a chat, so new conversations appear promptly even if realtime updates are delayed.

Email changes confirm reliably

Account email-change confirmation links now verify correctly, including setups where both the current and new inboxes need confirmation.

MiniMax and Kimi upgraded on Together AI

We updated the Together AI chat lineup in magicdoor.ai to MiniMax M3 and Kimi K2.7 Code with verified serverless model IDs, pricing, context windows, and legacy routing aliases.

• Routes new MiniMax chats to MiniMaxAI/MiniMax-M3 and new Kimi chats to moonshotai/Kimi-K2.7-Code
• Keeps legacy MiniMax M2.5/M2.7 and Kimi K2.5/K2.6 saved values resolving to the current endpoints
• Refreshes Together AI pricing, cached-input pricing, vision metadata, model pages, resources, and onboarding copy

Parallel chat streams

Chat streams now continue in the background when you switch chats or navigate away, so you can start multiple conversations and come back to find them finished.

• Active streams keep running when you switch to a different chat
• Switching back to a streaming chat shows live progress immediately
• Stop button only cancels the currently selected chat stream

Clearer upgrade prompts for trial users

Trial, billing, and locked-feature prompts now use one clearer upgrade flow across Chat, Images, Console, Usage, model selection, and customization.

• Routes trial users who run out of credit to Subscribe and active subscribers with low balance to Usage
• Uses friendlier upgrade prompts instead of warning-style low-balance messages
• Keeps pricing details on the Subscribe page so in-app prompts stay short

Chat streams now finish after disconnects

Chat responses now keep finishing server-side when a tab is closed, refreshed, or disconnected, so saved replies and usage tracking complete more reliably.

Claude Opus upgraded to 4.8

We updated magicdoor.ai to use Claude Opus 4.8 while keeping Opus pricing unchanged.

• Routes new Opus chats to claude-opus-4-8 and keeps legacy Opus 4.6/4.7 chats on the current model
• Keeps Opus pricing at $5 input and $25 output per 1M tokens, with the same cache pricing
• Refreshes model pages, resource guides, marketing copy, and onboarding copy to show Claude Opus 4.8

Oggie setup now self-registers

Magicdoor now creates and reuses one app-level Oggie OAuth client per deployment, so users can connect Oggie without a manually preconfigured client ID.

• Registers the Magicdoor deployment with Oggie automatically on first connect
• Reuses that app-level client across users while keeping each user OAuth grant separate
• Shows clearer account-page messages when Oggie setup is temporarily unavailable or server configuration is missing
• Classifies missing connection-token encryption setup as a Magicdoor config issue instead of a temporary Oggie outage

AI subscription calculator added

We added a public calculator for visitors comparing multiple AI subscriptions against magicdoor.ai usage-based pricing.

• Shows how stacking ChatGPT, Claude, Gemini, and Perplexity-style subscriptions can reach $60-80/month
• Frames magicdoor.ai around the $6/month base subscription, included $1 credit, and typical $8-10/month total
• Links high-intent visitors into signup, usage-based pricing, and supported model pages

Long chats stay smoother

We improved magicdoor.ai Chat performance for long conversations by only mounting the visible message window once a thread gets large.

• Keeps high-value long chats responsive as message counts grow
• Preserves existing chat controls like images, artifacts, compare, fork, costs, and code-interpreter state
• Reduces avoidable browser work before users hit scroll or typing lag

Oggie calendar connection foundation

We added the first secure foundation for connecting Oggie to Magicdoor so chat can receive server-side calendar tools when a user has connected their account.

• Adds an Oggie card to Account integrations with connect, reconnect, and disconnect states
• Stores Oggie connection tokens encrypted server-side and keeps decrypted tokens out of the browser
• Adds read-only Oggie calendar MCP tools for connected users

Chat can recover when a model is unavailable

When a chat model hits a temporary provider issue, magicdoor.ai now preselects a compatible fallback in the existing retry flow so you can continue the same prompt.

• Detects retryable model, capacity, timeout, and server errors
• Suggests only fallback models that match the prompt requirements
• Keeps recovery attached to the original retry card

Gemini image limit guidance added

We added a focused public guide for people who hit Gemini image generation limits and need to decide whether to wait, stay with Google, or try a usage-based image workflow.

• Explains when Google still makes sense for very heavy Gemini usage or Google-first workflows
• Shows how magicdoor.ai handles image work with multiple models, visible usage pricing, and no magicdoor.ai-imposed cooldowns
• Links directly into the image generator for users who want to keep creating

Removed ChatGPT conversation import

We removed an accidental ChatGPT conversation import surface from Account settings and the matching import endpoint.

• Removes the conversation import panel from Account settings
• Removes the export upload endpoint and parser
• Keeps migration guidance focused on manually copying saved memory and preferences

DeepSeek V4 Pro added to Chat

We added DeepSeek V4 Pro as a managed chat model in magicdoor.ai through the existing Together AI integration.

• Routes DeepSeek chats to deepseek-ai/DeepSeek-V4-Pro
• Adds DeepSeek pricing, context window, and conservative text-only attachment metadata
• Keeps Mistral out for now because current Together AI serverless pricing was not available in the verified pricing source

Installed Chat opens faster

We reduced the work that blocks the first magicdoor.ai/chat shell, so installed and bookmarked launches can show the chat surface sooner on mobile.

• Keeps auth and subscription guardrails in place before Chat opens
• Loads preferences, balance, renewal details, and sidebar history after the shell starts rendering
• Improves standalone launch polish with matching theme colors and safe-area padding

Homepage model labels refreshed

We refreshed the main magicdoor.ai homepage marquee so it shows the current GPT-5.5 and Claude Opus 4.7 labels instead of older model names.

• Updates the homepage marquee from GPT-5.1 to GPT-5.5
• Updates the Claude Opus label from 4.6 to 4.7
• Keeps the public landing page aligned with the current model registry

Find the right AI model before you start

A new public /which-ai-model page helps visitors choose a practical model for writing, coding, research, images, or everyday chat before signing up.

• Adds a four-question model recommender with one primary recommendation and alternatives
• Uses current magicdoor.ai model labels, capabilities, and usage pricing for cost posture
• Keeps the flow public and separate from future first-login onboarding

Kimi upgraded to K2.6

We updated the Moonshot AI chat model in magicdoor.ai to Kimi K2.6, adding the new Together AI endpoint and refreshed multimodal coding metadata.

• Routes new Kimi chats to moonshotai/Kimi-K2.6
• Keeps older saved Kimi K2.5 chats resolving to the new K2.6 endpoint
• Refreshes model labels, pricing, and marketing copy to show Kimi K2.6

Chats now know their active model

Magicdoor now includes the active model context in each chat generation prompt, so assistants can answer model-identity questions based on the model currently handling the message.

• Adds the active model ID to chat system context for each generation
• Includes model label and provider details when available
• Keeps assistant instructions, memory, and project instructions intact

Install magicdoor.ai from your phone

A new public /install page explains how to open magicdoor.ai/chat first, then add it to your home screen for faster access to Chat.

• Shows a dismissible mobile banner on Chat and Images that links to the install guide
• Directs mobile users to start from magicdoor.ai/chat before using Add to Home Screen
• Includes iPhone Safari and Android Chrome install steps
• Keeps the install guide publicly reachable at magicdoor.ai/install

Grok upgraded to Grok 4.3

We updated the xAI chat model in magicdoor.ai to Grok 4.3 while keeping the existing pricing and capability metadata in place.

• Routes new Grok chats to the grok-4.3 model ID
• Keeps existing saved Grok chats resolving to the new Grok 4.3 endpoint
• Refreshes model pages, resources, and product copy to show Grok 4.3

GPT-5.5 now leads the chat picker

We reordered the chat model picker so GPT-5.5 is the first default option for new chats, followed by Claude Sonnet 4.6 and GLM-5.1.

• Makes GPT-5.5 the first model new and trial users encounter
• Keeps Claude Sonnet 4.6 as the broadly available Claude option near the top
• Keeps GLM-5.1 available directly below Claude Sonnet 4.6

GPT reasoning now starts on Low

GPT chats now default to Low reasoning effort, while the Medium and High options remain available in the reasoning selector.

GPT-5.5 chat tools now finish reliably

GPT-5.5 chats now keep running after OpenAI tool results so Magicdoor tools and code interpreter sessions can complete with a final answer.

• Restores multi-step GPT-5.5 tool calls for search, memory, canvas, and other chat tools
• Keeps code interpreter progress and generated-file capture deterministic
• Persists code interpreter containers even when no generated file is cited

Chat onboarding now shows models and images together

New users now get a clearer chat onboarding flow that highlights model switching first, then includes one complimentary Seedream 4.5 image after they try a second model.

• Guides new users to compare models in the same chat
• Adds one complimentary Seedream 4.5 onboarding image
• Keeps future image generation clearly billed per image

GLM-5.1 now leads the chat picker

We reordered the chat model picker so GLM-5.1 is the first default option for new chats, with Gemini 3.1 Pro and Claude Haiku moved lower.

• Makes GLM-5.1 the first model new and trial users encounter
• Keeps Gemini 3.1 Pro and Claude Haiku available lower in the picker
• Preserves MiniMax M2.7 and legacy M2.5 routing

MiniMax upgraded to M2.7

We replaced the deprecated MiniMax M2.5 endpoint with MiniMax M2.7 across magicdoor.ai after the provider sunset notice.

• Routes new MiniMax chats to MiniMaxAI/MiniMax-M2.7
• Keeps older saved M2.5 chats resolving to the new M2.7 endpoint
• Refreshes model labels and resource copy to show MiniMax M2.7

Long chats now reuse cached prompt context

Prompt caching is now enabled for Claude and GPT chats, lowering cost and latency when longer conversations reuse the same context.

• Claude system context is marked cacheable for supported long chats
• GPT chats use account-scoped prompt cache routing
• Cached token reads and writes are tracked separately for accurate billing

Compare image models from one picker

The /images flow now lets you choose one model for a normal generation or select multiple compatible models to compare outputs side by side.

• Model rows generate with a single model, while checkboxes add models for comparison
• Aspect ratios stay compatible with the selected model set
• Compare results show model-labeled image tiles with per-model loading and error states

ChatGPT Image upgraded to gpt-image-2

We upgraded the ChatGPT Image path in magicdoor.ai to Replicate gpt-image-2 and aligned the images flow around the newer model.

• Updated the main image route and library plumbing for the upgrade
• Refreshed image model labels across the product
• Added route and image utility test coverage for the change

File context is now visible above the composer

Chats now show which files and images are currently in model context, with detach controls when you want to narrow the prompt.

• Historical attachments stay visible in the context bar
• Historical image thumbnails now render cleanly
• Non-vision TogetherAI models safely strip file parts before requests

Image editing mode landed in /images

The /images page gained a dedicated editing mode for uploaded images, plus signed uploads and library updates to support the flow cleanly.

• Edit existing images from the main images workflow
• Signed URL support added for image uploads
• Image form and library UI updated for the new path

Privacy and security got a public home

We launched a public trust page covering data handling, account-scoped conversations, and product privacy details in plain language.

• Public /trust landing page is now live
• Canonical metadata and structured data added
• Middleware updated so the page stays publicly reachable

Comparison pages became scalable

magicdoor.ai got a dynamic /compare system backed by structured data, making it easier to ship and maintain head-to-head pages.

• Started with 6 competitor pages
• Reusable comparison table component
• Per-page SEO metadata and sitemap coverage

PWA startup got faster on repeat visits

We shipped cold-start quick wins for the PWA, including better app-shell caching, deferred scripts, and lighter initial chat loading.

• Improved service worker caching for /chat
• Deferred the markdown fix script until after interactive
• Dynamic imports reduced the upfront chat payload

Added Seedream 4.5 for image generation and editing

Seedream 4.5 joined the image lineup at $0.03/image with support for both generation and single-image editing inputs.

• Added bytedance/seedream-4.5 to the image model list
• Created inputs for prompts, aspect ratio, and edit images
• Placed alongside Nano Banana Pro and Imagen 4 in the lineup