Changelog

Features

Wallet Revamp

• Redesigned Wallet page with clearer Credits breakdown display
• Added payment history section for transaction tracking
• Access via Wallet

Anthropic Models API

• New endpoint: https://zenmux.ai/api/v1/anthropic/models
• Query models supporting Anthropic Messages protocol
• Returns filtered model list with protocol compatibility

Holiday Theme

• Launched Christmas theme with refreshed UI
• Seasonal design updates across platform

UX Improvements

• Improved scrollbar styling for better visual consistency
• Refined Studio-Chat system prompt handling logic
• Improved Studio-Chat PK button display logic

Fixes

• Fixed icon display bug on Cost and Usage pages

API Change Notice

Free model slugs are being standardized with a -free suffix to distinguish them from capacity-guaranteed paid models.

Effective Date: January 1, 2026

Affected Models

The following free model slugs are changing:

Action Required

• Update API calls to use new slugs with -free suffix
• Old slugs will continue to work during transition period
• All future free models will follow the -free naming convention

New Model

GLM 4.7

• GLM 4.7 is now available on ZenMux
• Improved performance across reasoning, coding, and instruction tasks
• Access via model page

New Model

MiniMax M2.1

• MiniMax M2.1 is now available on ZenMux
• Advanced multimodal understanding and generation
• Access via model page

New Model

Doubao-Seed-1.8

• Doubao-Seed-1.8 is now available on ZenMux
• Enhanced reasoning and instruction-following capabilities
• Access via model page

New Model

Gemini-3-Flash-Preview

• Gemini-3-Flash-Preview is now exclusively available on ZenMux
• Free tier: Limited usage in Studio-Chat, refreshes in UTC 0:00 daily
• Paid tier: Unlimited access via API and Studio-Chat
• Access via model page

New Model

Xiaomi MiMo-V2-Flash

• MiMo-V2-Flash is now available on ZenMux
• Free to use for all users
• Access via model page

UX Improvements

• Improved invite code interaction and display UI
• Improved invite code input experience during user registration
• Refined Studio-Chat sidebar display logic
• Improved dark mode usability and visual consistency

Fixes

• Fixed user account status not syncing correctly
• Fixed Studio-Chat page stuck on the loading state and becoming unusable

New Models

OpenAI GPT-5.2

• GPT-5.2 is now available on ZenMux
  • Access via model page

OpenAI GPT-5.2 Chat

• GPT-5.2 Chat is now available on ZenMux
  • Access via model page

OpenAI GPT-5.2 Pro

• GPT-5.2 Pro is now available on ZenMux
  • Access via model page

Features

Nano Banana Pro Google Vertex Protocol Update

• Adapted Nano Banana Pro (Gemini 2.5 Flash Image) to Google Vertex AI latest protocol
• Added Signature support for enhanced image consistency
• Improved accuracy in image editing scenarios
• Delivers significantly better cross-request visual coherence

Improvements

Models Page Advanced Sorting

• Sort models by latency on models page
• Sort models by throughput for performance comparison
• Quickly identify fastest models for your use case

Studio-Chat User Experience

• Input field remains editable during content generation. Send button disabled while streaming to prevent duplicate requests
• Sidebar chat button creates new conversation instead of resuming previous session

Action Required

The following limited-time free model endpoints will go offline in 1 hour:

• Gemini 3 Pro Image (Nano Banana Pro) Free
• Gemini 3 Pro Preview Free
• Gemini 2.5 Flash Image (Nano Banana) Free
• Gemini 2.5 Pro Free

Required Action

Switch to standard paid endpoints immediately to avoid service interruption.

GLM 4.6V Series Models

Developers can now integrate GLM 4.6V models through ZenMux endpoints.

• Access GLM 4.6V: GLM 4.6V
• Access GLM 4.6V Flash (free tier): GLM 4.6V Flash

Features

Automatic Recharge (Auto Top-Up)

• Configure automatic balance recharge in wallet settings
• Access via "Auto Top-Up" button in top-right wallet menu
• Set custom thresholds to maintain account balance

Dark Mode

• Enable dark mode in console settings
• Optimized for low-light environments
• Reduces eye strain during extended sessions

Improvements

Models Page URL Parameters

• Navigate directly with query parameters
• Filter and sort models via URL (e.g., https://zenmux.ai/models?series=GPT&sort=newest)
• Bookmarkable filtered views

Studio-Chat Streaming for Nano Banana Pro

• Gemini-3-Image-Pro-Preview (Nano Banana Pro) switched to streaming mode
• Real-time response delivery for improved interaction experience

Enhanced Logs Billing Details

• Console logs display detailed billing breakdown
• View individual cost items per request
• Improved transparency for usage tracking

Mistral Large 3 Model

Developers can now integrate Mistral Large 3 through ZenMux endpoints.

• Access the model: Mistral Large 3

DeepSeek V3.2 Series Models

Developers can now integrate DeepSeek V3.2 models through ZenMux endpoints.

• Access DeepSeek V3.2 official release: DeepSeek Chat and DeepSeek Reasoner
• Access DeepSeek V3.2 Speciale edition: DeepSeek V3.2 Speciale

Claude Opus 4.5 Rollout

Developers can now integrate the Claude Opus 4.5 through ZenMux endpoints.

• Access the model: Visit Claude Opus 4.5

Invite Registration

Users can now invite others to register for ZenMux.

• Share your invite link to earn rewards
• Find your unique invite link in Account Settings

Studio-Chat Image Generation Settings

Studio-Chat now supports granular image generation controls.

• Set output resolution for generated images (1k, 2k, 4k, etc.)
• Choose aspect ratio (1:1, 16:9, 4:3, etc.)
• Settings panel available in the image generation chat interface

Nano Banana Pro Enhancements

Nano Banana Pro model gains new capabilities.

• Web Search support added for real-time information retrieval
• Image generation API switched to non-streaming for improved stability

Improvements

• Clearer discount info display on recharge button
• Studio-Chat interaction experience refinements
• Studio-Chat mobile experience optimizations
• UI detail polish

Fixes

• Fixed issue causing frequent 400 errors

Nanobanana Pro (Gemini 3 Pro Image Preview) Arrives

Visual generation just leveled up! We are absolutely thrilled to announce that Nanobanana Pro (Gemini 3 Pro Image Preview) is finally released and available first on ZenMux.

The results are explosive. Is this the new standard for AI imagery? Don't just take our word for it—witness the mind-blowing effects yourself!

• Next-Gen Imaging: Developers can integrate the stunning Nanobanana Pro capabilities immediately.
• Accessible to All: We've launched a FREE VERSION so you can start creating right now (subject to rate limits).
• Premium Performance: Upgrade to the paid tier for a DEDICATED LINE, ensuring silky-smooth usage without interruptions.

Grok 4.1 Fast Series Models

• You can now access the new Grok 4.1 Fast model and Grok 4.1 Fast(Non- Reasoning) model through ZenMux AI platform.

Google Gemini 3.0 Pro Preview Unleashed!

The benchmark hasn't just moved—it's been shattered! We are absolutely thrilled to announce that Gemini 3.0 Pro from Google DeepMind is finally here on ZenMux.

Is this the new King of AI? Does it crush the competition? Stop reading the charts and start witnessing the power yourself!

• Experience the New SOTA: Developers can integrate the full Gemini 3.0 Pro lineup immediately.
• Exclusive Free Tier: We've launched a DEDICATED FREE VERSION (with daily quotas) so you can run your own evals right now.
• Instant Access: Start benchmarking immediately via ZenMux API calls and Studio-Chat.

New Free Models

Users can now run Gemini-2.5-Pro and Gemini 2.5 Flash Image (Nano Banana) without cost, within rate limits.

Studio-Chat

Studio-Chat now lets you switch to free models and warns when you hit their limits.

API

API endpoints accept the new free models and enforce rate controls.

Fixes

• Restored GPT-Codex compatibility with the Anthropic protocol, so Codex-based apps can resume Anthropic-compliant calls

OpenAI GPT-5.1 Series Models Rollout

Developers can now integrate the full GPT-5.1 lineup through ZenMux endpoints.

• Access GPT-5.1, GPT-5.1 Chat, GPT-5.1-Codex, and GPT-5.1-Codex-Mini immediately via API calls and Studio-Chat

ERNIE 5.0 Thinking Preview Model

• You can now access the new ERNIE 5.0 Thinking Preview model through ZenMux AI platform. Visit ZenMux ERNIE 5.0 Thinking Preview Model for more information.

VolcanoEngine: Doubao-Seed-Code Model

Users can now access ByteDance's latest programming model through ZenMux.

• New model: VolcanoEngine: Doubao-Seed-Code
• Specialized for programming tasks
• Available in Studio-Chat and API endpoints

Studio-Chat Style Improvements

Enhanced the visual and interactive experience of Studio-Chat.

• Optimized reasoning process display styles for better readability
• Unified error notifications to appear within chat boxes for consistent user experience
• Improved Auto Router model selection UI for easier model choice
• Fixed input lag when uploading attachments and typing text simultaneously

Bug Fixes

Resolved issues affecting model PK functionality.

• Fixed synchronization switch not working in Model PK mode
• Ensured real-time synchronization across PK chat windows when the feature is enabled

API Key Management

Users can now create up to 20 API keys per account.

• Maximum limit of 20 API keys per user
• Prevents excessive key generation
• Helps manage API usage more effectively

Homepage Redesign

The ZenMux homepage has been completely redesigned with a fresh, modern interface.

• Clean, updated visual design
• Improved navigation and user experience
• Faster loading and better performance
• Modern layout showcasing key features

Studio-Chat Image Editing Skills

Studio-Chat now includes advanced image editing capabilities in the Octopus Skills mode.

• AI Background Removal: Automatically remove backgrounds from images with one click
• AI General Editing: Comprehensive image editing tools powered by AI
• Integrated seamlessly into the Studio-Chat interface
• Available for all supported image models

Chat Page Reasoning Enhancements

Models with toggleable reasoning capabilities now have improved default settings.

• Thinking mode enabled by default for applicable models
• Default reasoning token value set to 1024
• Reduces manual configuration for reasoning tasks
• Optimized for better reasoning performance out of the box

Kimi K2 Thinking Model

• You can now access the new Kimi K2 Thinking model through ZenMux AI platform. Visit ZenMux Kimi K2 Thinking Model for more information.

Qwen3 Max Thinking Preview Model

• Users can now access the new Qwen3 Max Thinking Preview model through ZenMux AI platform. Visit ZenMux Qwen3 Max Thinking Preview for more information.

Frontend Improvements

Studio-chat interface now provides a cleaner, more intuitive visual experience with optimized loading states.

• Studio-chat Visual Optimization: Improved chat interface layout and styling for better readability
• Global Loading Optimization: New octopus-themed loading interface provides visual feedback during operations
• Bug Fixes: Various UI and interaction issues resolved for smoother user experience

Backend Enhancements

ZenMux now supports more AI models with improved parameter compatibility and new image processing capabilities.

• x.ai Model Compatibility: Enhanced parameter handling for non-standard x.ai model configurations
• Static Resource Optimization: Fixed cross-origin issues for better resource loading
• Ming Model Expansion: Added support for additional effect parameters and Vertex AI Imagen API integration
• Image Processing Features: Ming model now supports image generation, editing, and background removal
• Vertex AI Integration: Full adaptation to Google's Vertex AI Imagen API for advanced image operations

Image Matting (beta)

Ming model can now remove backgrounds from images with high precision.

• Background Removal: Automatic image matting functionality for clean subject isolation
• Available via API: Access image matting through the Ming model endpoints
• High Quality Results: Professional-grade background removal for various image types

Google Vertex AI Native Support

ZenMux now supports Google's native Vertex AI protocol for maximum compatibility:

• Native Protocol: Use Vertex AI's original API format alongside OpenAI/Anthropic protocols
• Full Feature Parity: Access all Vertex-specific features including multimodal inputs
• Automatic Version Support: v1 and v1beta API versions supported

Full Multimodal Capabilities

ZenMux now supports all major content modalities across supported models:

• Audio Input/Output: Voice conversations and audio generation
• Image Processing: Image understanding, generation, editing, and upscaling
• Multimodal Mixing: Combine text, images, audio, and video in single requests
• All multimodal features automatically metered with transparent pricing

Studio-Chat Enhancements

Major improvements to the interactive chat experience:

• Audio Playback: Play generated audio directly in chat interface
• Large File Optimization: Automatic URL-based handling for large audio/video files
• Better Loading States: Unified octopus loader across all generation types

Activity & Logs Improvements

Comprehensive overhaul of request monitoring and analytics:

• Advanced Filtering: Filter by model, provider, status, time range, and more
• Web Search Visualization: See web search costs and results inline
• Function Calling Display: Rich rendering of tool calls and responses
• 90-Day Retention: Updated data retention policy (documented in privacy policy)
• Model Quick Jump: Click any model to view cost analytics filtered by that model
• Detail view no longer refreshes when returning from inspection

Insurance & Compensation System

• Email Notifications: Instant alerts when compensation credits are issued
• Detailed Reports: High-latency and unsatisfactory content tracking
• Algorithm Improvements: Better detection of legitimate quality issues

Model Display Optimizations

Improved model discovery and information architecture:

• Tiered Pricing Highlight: Clearly marked volume-based pricing tiers
• Output Modality Filters: Filter models by output type (text, image, audio, video)
• Reasoning Type Filters: Filter by inference type (standard, thinking, chain-of-thought)
• Protocol Filters: Find models by supported protocols (OpenAI, Anthropic, Vertex)
• Provider Logos: Visual provider identification in model details
• Release Dates: See when each model was launched
• Full Token Limits: Display complete context window sizes
• Price Display: Upfront pricing information on model cards
• Back Button: Easy navigation from model details to list view
• Simplified Code Examples: Streamlined integration samples

Documentation Expansions

Major additions to help developers integrate and optimize:

• Vertex AI Guide: Complete tutorial for using Ming models via Vertex API
• Claude Code Custom Models: Instructions for subagent and custom model scenarios
• Cost & Pricing Guide: Comprehensive pricing and billing documentation
• Model & Provider Intro: Deep dive into model capabilities and providers
• Model Routing: Smart routing and fallback strategies
• Usage & Cost Observability: Analytics and monitoring best practices
• OpenChat Integration: Exploring custom provider configuration support

API & Infrastructure

Backend improvements for reliability and developer experience:

• CORS Support: Cross-origin requests enabled for web-based integrations
• Reasoning Field Support: Compatible with MiniMax M2 reasoning format
• Cloudflare KV Sync: Configuration data synced to edge for low-latency access

Anti-Abuse Measures

Stronger protections against service abuse:

• Zero-Balance Restrictions: High-cost models disabled for accounts with ≤$0 balance
• Friendly Error Messages: Clear guidance when models are restricted
• Studio-Chat Warnings: In-app notifications when attempting to use restricted models

Notification System

Comprehensive email notification framework:

• Recharge Alerts: Notifications when credits are added to account
• Compensation Notices: Automatic emails when quality compensation is issued
• Bonus Credits: Alerts for promotional credit grants

Strategy & Routing

Smart model selection and optimization features:

• Speed vs. Price Priority: Configure whether to optimize for latency or cost through console/strategy page

UX Improvements

• Improved loading animations for data-heavy pages and empty-state screens: smoother skeleton placeholders, unified empty-state visuals, and subtle entrance animations to reduce perceived wait time.

Fixes

• Fixed Gemini 2.5 Pro reasoning parameter validation errors
• Fixed Gemini 2.5 Flash-lite tool_choice not triggering function calls
• Fixed MiniMax M2 token detail fields returning None (cache/reasoning usage)
• Fixed MiniMax M2 developer message handling
• Fixed MiniMax M2 system message array format consumption
• Fixed MiniMax M2 assistant message array not working in reasoning
• Fixed MiniMax M2 reasoning_effort parameter rejection
• Fixed Imagen 3.0 chat interface stuck in loading state
• Fixed Grok-4 cache statistics showing incorrect 680-token values
• Fixed Ming cancelled status inaccuracy
• Fixed chat page losing conversation history when artifacts enabled
• Fixed chat sidebar names auto-updating despite manual edits
• Fixed chat input code formatting (newlines preserved)
• Fixed PK mode message sending while other windows still generating
• Fixed log detail page showing empty content before refresh
• Fixed API key usage display lag vs. admin backend
• Fixed copy button requiring double-click (expanded hit area)
• Fixed filter selection requiring reset instead of click-to-deselect
• Fixed Vertex stream_generate_content 500 error with candidate_count
• Fixed generationTime reporting as 0
• Fixed duplicate icons in UI
• Fixed text truncation/overlap issues
• Fixed loading from historical audio conversations
• Fixed security vulnerabilities
• Fixed Models page search functionality
• Fixed wallet → credit terminology inconsistency in logs
• Fixed content ending prematurely mid-response_

OpenCode Integration

You can now use ZenMux models directly in OpenCode, see more in Guide to Using OpenCode with ZenMux

Claude Code Integration

You can now use ZenMux models directly in Claude Code with minimal setup:

• Update base_url and API key in Claude Code settings
• Access all ZenMux-supported models with Anthropic protocol compatibility
• Full documentation available at docs.zenmux.ai/best-practices/claude-code.html_

Codex Integration

You can now use ZenMux models directly in Codex with minimal setup:

• Update base_url and API key in Codex settings
• Access all ZenMux-supported models with OpenAI protocol compatibility
• Full documentation available at docs.zenmux.ai/best-practices/codex.html_

Studio-Chat Artifacts

Generate and render interactive HTML artifacts directly in Studio-Chat:

• Activate artifact mode to enable HTML generation
• Models create self-contained HTML artifacts with live rendering
• Perfect for code demonstrations, visualizations, and interactive content

Studio-Chat Model PK

Compare multiple models side-by-side in Studio-Chat:

• Create parallel conversation windows with different models
• Run identical prompts across multiple models simultaneously
• Analyze responses, performance, and output quality side-by-side

Change Log Page Launch

ZenMux has launched its official Change Log Page, allowing users to stay informed about all platform updates in one place:

• View all version releases, feature updates, and improvements in real time

• Access directly from the top navigation bar or at changelog.

MiniMax-M2 Model Access

Use the new MiniMax-M2 model through ZenMux AI platform.

• Access the model: Visit ZenMux MiniMax-M2
• Try for free: Complimentary API calls available
• Limited time offer: Ends November 7, 2025 at 00:00 UTC

Gemini 2.5 Flash Image (Nano Banana) Model

Access the new Gemini 2.5 Flash Image (Nano Banana) model through ZenMux AI platform.

• Visit the model page at zenmux.ai/google/gemini-2.5-flash-image to learn more
• Available immediately for all users

Google Vertex Protocol Support (Beta)

ZenMux now supports Google Vertex protocol with initial capabilities:

• Generate content using Google Vertex's native protocol
• Currently supports text and image generation

Studio-Chat Enhancements

Enhanced chat interface with improved input/output flexibility:

• Upload images as chat inputs with model-specific file filtering
• Choose output modality (text or image) for each conversation
• View generated images directly in the chat interface
• Model cards now display input/output modalities and availability status
• Web Search uses each model's native capabilities when supported
• Web Search button automatically enabled/disabled based on model support

KAT-Coder-Pro-V1 Model

• Users can now access the new KAT-Coder-Pro-V1 model through ZenMux AI platform. Visit ZenMux KAT-Coder-Pro-V1 for more information.

Claude Hiku 4.5 Model

Claude Hiku 4.5 is now available on ZenMux, Visit ZenMux Claude Hiku 4.5 for more information.

Studio Chat File Uploads

The chat interface now supports direct file uploads for enhanced AI interactions.

Text Documents:

• .txt, .md, .markdown

Data Formats:

• .csv, .json, .xml

Programming Languages:

• .py, .js, .ts, .java, .c, .cpp, .h, .cs, .go, .php, .rb, .swift, .sql

Web & Configuration:

• .html, .htm, .yaml, .yml, .ini, .sh, .css

Upload files directly in the chat interface to enable AI analysis and processing.

Ring-1T Model

• Users can now access the new Ring-1T model through ZenMux AI platform. Visit ZenMux Ring-1T for more information

Alipay Payment Support

• ZenMux now accepts Alipay for wallet payments, expanding payment options for users.

Ling-1T Model Integration

• Users can now access the Ling-1T model through ZenMux, Visit ZenMux Ling-1T for more information

Initial Release

Welcome to ZenMux - the world's first insurance-backed LLM aggregation platform. We're solving the real challenges developers face: fragmented provider ecosystems, quality uncertainty, and cost unpredictability.

Unified Model Access

Access all major LLM providers through a single API key and unified interface.

• Create one API key to access OpenAI, Anthropic, Google, DeepSeek, and more
• No need to register on multiple platforms or manage multiple wallets
• Seamless model switching without code changes
• Unified billing and transparent cost tracking

Dual-Protocol Support

Choose the API protocol you're most comfortable with.

• OpenAI-compatible API: Use OpenAI's standard API to invoke any model on the platform
• Anthropic-compatible API: Use Anthropic's standard API with seamless Claude Code integration
• Switch protocols without changing your model provider
• Zero learning curve if you're already familiar with either API

High Availability

Enterprise-grade reliability with multi-provider redundancy.

• Nearly all models provisioned at Tier 5 capacity quotas
• Automatic failover when provider capacity is saturated
• Multi-layered capacity reserves across providers

Transparent Quality Testing

Regular degradation checks across all models, with results open-sourced on GitHub.

• Human Last Exam (HLE) tests run regularly for all models on all channels
• Approximately $4,000 invested per test run
• Complete testing process and results publicly available on GitHub
• Real-time quality leaderboard at zenmux.ai
• Ensures all models are authentic and free from degradation

World's First AI Model Insurance

Automatic payouts for Unsatisfactory Content, and High Latency.

• Comprehensive coverage for unsatisfactory outputs and high latency
• Automated daily detection system
• Payouts automatically credited the next business day
• High-quality bad case data provided for product optimization
• No additional setup required

Intelligent Model Routing

Automatically select the best model for each request to balance quality and cost.

• Analyzes request content and task characteristics
• Routes to optimal model without manual selection
• Detailed routing decision logs for transparency
• Support for custom routing rules
• Continuous learning from historical data

Comprehensive Observability

Complete visibility into your AI application's performance with built-in analytics dashboards.

• Detailed Logs: Full request and response details for every API call
• Cost Analytics: Multi-dimensional cost analysis by project, model, provider, and time
• Usage Tracking: Real-time token consumption and call frequency monitoring
• Performance Metrics: Response time, latency, concurrency, and throughput tracking
• Model Comparison: A/B testing support for evaluating different models
• Visual Dashboards: Built with Next.js 15, React 19, and Recharts
• 119-field analytics model tracking 10 billing types across all time dimensions

Global Edge Network

Low-latency access worldwide powered by Cloudflare infrastructure.

• Distributed edge nodes across all continents
• Automatic routing to nearest available node
• Consistent performance regardless of location
• Highly available multi-node architecture

Resources & Documentation

• Product Website: zenmux.ai
• Claude Code Integration Guide: docs.zenmux.ai/best-practices/claude-code.html
• CodeX Integration Guide: docs.zenmux.ai/best-practices/codex.html
• Discord Community: discord.gg/vHZZzj84Bm
• Twitter: @ZenMuxAI
• Technical Support: [email protected]
• Business Inquiries: [email protected]

What's Next

We're building ZenMux with a developer-first approach. This v1.0.0 release is just the beginning - we're committed to continuous improvement based on your feedback.

Your feedback shapes ZenMux. Join our Discord community or reach out to [email protected] to share your thoughts.

ZenMux: With a Zen mindset, harness the power of AI — countless models unified in one place, achieving optimal results through the simplest experience.