What's new in Big-AGI.

Chats list: improved looks and menu, and up/down keyboard navigation

Reliability and security fixes for logins, browser cache notices, and Openrouter/OpenAI protocols

OpenAI GPT-5.6 Pro reasoning mode and Max effort, also via OpenRouter

Anthropic prompt caching on OpenRouter

Gemini Nano Banana 2 Lite

SpaceXAI Grok 4.5 support with 500k context window and effort adjustment

Gemini Omni Preview support with short video generation

Cohere models family support

Anthropic Sonnet 5 support

Cerebras ultra-fast inference support: Gemma 4 31B, GPT-OSS 120B, GLM 4.7

Added daily model and provider rankings

OpenAI GPT-5.6 model family with Sol (coding, science, cybersecurity), Terra (balanced), Luna (faster, cheaper)

Alibaba models including Qwen3.7 Max, plus Alibaba-served DeepSeek V4 Pro, GLM-5.2, and Kimi K2.7 Code

Sakana AI official support including Fugu and Fugu Ultra models

Improved resilience to storage issues

New Settings interface

Models: new models indicators, extend OpenRouter reasoning controls

Better controls for Anthropic code execution

Gemini search citations resolve to the actual link rather than an opaque redirect

Pro users can recover deleted chats with the cloud trashcan

Remove older registered devices from the account page

Z.ai GLM-5.2 and Moonshot Kimi K2.7-code support

Note: the access to Fable 5 is still disabled

Anthropic Claude Fable 5 and Claude Mythos support with 1M context

Scratch Chats: search toggle and materialize into full chats

Beam Merge: editable separators, with custom editor

OpenAI Code Interpreter: download files created in the sandbox

Gemini code execution and Deep Research files download

Markdown-within-markdown code rendering support

Beam and Merge support resumable Gemini Deep Research agents

Gemini Deep Research and Antigravity Agent models support via new Interactions API

MiniMax M3 support including image inputs, and 1M context

Anthropic reasoning token accounting, and Merge improvements

Claude Opus 4.8 support, and Grok Build 0.1, and stable Gemini Nano Banana 2 and Pro

Auto model selection: best model per task, cross-vendor

Re-beam all merges, partial removal, fixed thinking block stripping

Gemini 3.5 Flash and Gemini Antigravity Agent models with linux sandbox support

xAI models updated with none and x-high reasoning effort levels

Large improvements to Merges, especially of complex responses with reasoning/images

Open 2.0.5 - “Roberto” stable on GitHub and Docker (850K downloads)

Fixed xAI reasoning summaries, Anthropic execution continuity, OpenAI Responses sealed-stream

Models list shows publication dates and flags new models

Gemini 3.1 Flash Lite and Grok 4.3 support

Anthropic dynamic filtering disabled by default, active in “Search Max” mode

OpenAI GPT-5.5 model family with 1M context, and a no-thinking variant for temperature control

DeepSeek V4 Pro and V4 Flash with 1M context and max effort support

Resume and stop controls for interrupted Deep Research generations

Gemini Deep Research Max models with resume for interrupted answers

OpenAI GPT-Image-2 for image drawing and chat image generation

Model updates including Moonshot Kimi K2.6

Custom Persona Merges, AI Inspector improvements

Claude Opus 4.7 support with 1M context tokens and the new X-High effort

Create Images, PDFs, DOCs, and more with GA Anthropic Skills support

Gemini 3.1 Flash TTS and Robotics-ER 1.6 models

Anthropic Files now embed inline

Mad optimizations: code rendering much faster on large streams

Improved chat search

Anthropic file downloads in chat, persistent containers for Skills and bash sessions

New vendors: MiniMax including M2.7, and Arcee AI with Trinity models

Gemma 4 models and refresh to Groq and LocalAI

Anthropic Dynamic Filtering now default for Search mode

Improved chat search results relevance

Beam: re-beam merged responses

AI tool operations status display with nested trees

OpenAI code interpreter image outputs

Desktop notifications for AI responses

Open 2.0.4 stable on GitHub and Docker (820,000 downloads)

Anthropic dynamic web filtering and continuation of extra-long outputs

Max Mode in Beam: one click to push all models to maximum thinking and reasoning efforts

xAI Grok 4.20 and OpenAI GPT-5.4 mini/nano models

Improved Beam merge editor, added Beam override indicators, bubble menu redesign

Read aloud controls and AI recap: persona speaks short message summaries

Rich text copy fix, auto-title crash fix

Persona chat reinforcement instructions to optionally maintain model behavior

Live controls: reworked audio output and camera input, and fixes to minimaps and images

Lossless images support, e.g. 4K Nano Banana 2 produces 24MB files

OpenAI GPT-5.4 and GPT-5.4 Pro support with 1M context and reasoning controls

OpenAI GPT-5.3-Instant, Gemini Nano Banana 2, and Anthropic search depth controls

Attachments on any message, persona starters, mobile focus mode

AWS Bedrock support with native Anthropic, Amazon Nova and OpenAI-compatible models

OpenAI GPT-5.3-Codex and LLMAPI.ai support and mobile UX improvements

Anthropic Sonnet 4.6 and Google Gemini 3.1 Pro models support

Anthropic Fast mode option to run Opus 2.5x faster at 6x the cost

Added Z.ai models including GLM-5, with configurable thinking

Anthropic Opus 4.6 with adaptive thinking, Max effort, 1M tokens, and US-inference lock

Quick launch of Beam Teams, and added option to discard reasoning traces

OpenRouter model parameters are state of the art

Open 2.0.3 stable on GitHub and Docker (800,000 downloads)

Attach files from Google drive, with optimal LLM conversion

Added Inworld text to speech support with streaming playback

Model duplication and reorder messages in cleanup mode

Improved compatibility with LiteLLM and other gateways

Support for Kimi K2.5 comes with image understanding and reasoning

Google Docs-compatible pasting of responses, improve AI Inspector

xAI upgrade allows to search X posts, search the web, run code

Simplify Text or HTML attachments and convert to Markdown for higher signal

Option to allow unlimited text-to-speech time, defaults to ~3 minutes

Support for OpenAI osb-120b, not announced yet

Added collapsible model groups, AI Inspector on direct connections

Fixed direct speech, reasoning blocks order, table rendering with breaks

Improve configuration of OpenAI-compatible custom AI services

Novita.ai models support with full capability detection

Open: community fixes for LLM parameters, service renames

OpenAI GPT 5.2 Codex with x-high reasoning and native search

PDF Scans: auto-OCR or attach as images

Improved Persona editing (image, emoji, voice) and CJK input

Open: fix for Anthropic breaking API changes and OpenAI keepalives

OpenRouter image generation with Gemini Nano Banana Pro and more

Mobile UX fixes for popups drag/interaction

Gemini 3 Flash support with 4-level thinking: high, medium, low, minimal

OpenAI GPT Image 1.5 support

Happy New Year 2026! We can’t wait to get started, and thanks everyone for your support!

Happy Holidays everyone! We’ll be back in 2026, Jan 1st, with the top AI assistance for you

Accommodate Gemini API changes for ‘medium’ thinking levels

OpenAI GPT-5.2, and GPT-5.2 Pro, and GPT-5.2 Instant including the new X-High thinking level

Open security fixes for CVE-2025-55182, and to prevent injections in the GitHub AI triage system

Paste improvements: preserve rich structure while removing unwanted background colors

Improvements to Persona grids and dialogs and to Free/Pro minimaps

Note: updates will slow down during Holiday weeks, and will be back with Prompts at the end of the year

GPT 5.1 Codex Max support

Deepseek V3.2 Speciale comes with almost Gemini 3 Pro performance but 20 times cheaper

Mistral 3 family models, including Large 2512

New text to speech & Voices: support for LocalAI, OpenAI, WebAudio (free) in addition to ElevenLabs

Direct connection now available for all AI Services

Anthropic Opus 4.5 with controls for effort (speed tradeoff), thinking budget, search, etc.

Direct connection to cloud and local AI services from the browser for Anthropic, Google, LocalAI

Thanksgiving promo: $10.99/mo locked forever, will increase after

Released Open 2.0.1 on GitHub and Docker

UX changes: interleaved model thoughts, starred models display, persona starters, ai inspector fields

Maximum privacy with option to connect directly from the browser to Anthropic, Gemini, and LocalAI

Nano Banana Pro support including search, editing with up to 5 people/object references, image sizes and ratios

Grok 4.1 Fast support with automatic code execution and search

Voice calls save the transcripts

Gemini 3 Pro support including Browsing, Code execution, Thinking depth, Media resolution

Image generation (/draw) with Azure and LocalAI

Performance improvements

OpenAI GPT-5.1 model family

Social coding: “Improve Big-AGI“ support button for free Big-AGI feature development

Advanced support for OpenRouter including reasoning, search, multimodal

Improved OpenRouter integration with direct login

Added Kimi K2 Thinking and more Moonshot AI models

Reduced connection issues to overloaded AI servers by 85% thanks to 3 error correction methods

Generate Persona images

Chat chimes and spoken briefings - enable in chat options

Fixes to login links with Outlook, added AI image attachment captioning, nanosecond-level optimizations

Search toggle for beams

Option to manually set per-model pricing

Login with email, via magic link

Released Open 2.0.0 on GitHub and Docker

Resilience: Auto-retry LLM calls when upstream servers are temporarily unavailable

Search toggle to quickly turn on/off search for models that support it

Change context window size for custom models

Anthropic Search & Fetch support across models, making it the 5th provider with native search support

Anthropic Sonnet 4+ models support 1M Token context windows

Stability improvements, fixing Perplexity models and keyboard navigation

Anthropic Haiku 4.5 models (standard and thinking variants)

OpenAI GPT-5 Search API model for web-integrated responses

Updated models across Anthropic, Mistral, Deepseek, Groq, xAI, Perplexity

OpenAI Audio and Audio Mini models for audio output

Gemini models support Google Search

xAI Grok 4 Fast Reasoning and Grok Code Fast 1

AI Request Inspector with quick toggle, for maximum transparency

Mobile UI dialogs improvements

OpenAI GPT-5 Pro and gpt-image-1-mini

Cloud Sync launched! - long awaited and top requested, available now on app.big-agi.com

Dark launch - big thanks to the early testers

Anthropic Claude Sonnet 4.5 support

OpenAI GPT-5 Codex support

Gemini 2.5 Flash Preview and Robotics-ER 1.5 support

AI experts: view the exact AI API requests with the “AI Inspector” tool

Address the next message to a specific @Persona (beginning of the message only)

Beam: auto-select 4 good models at any time

Persona: deep data comparison with live editing

Improve personas image data support

GPT-5 Image Generation: look for the option on the side pane. Also improved Search, Progress indication, Citations.

Model controls and selection from the side panel

Personas now support Image data, perfect for designer personas, especially with Gemini Nano Banana

Drag to reorder personas in any list

… Former release note entries are removed for space

Welcome to the new Big-AGI, this will feel like a brand new product. If you are here for the first time, welcome, and I hope you’ll love the obsession and mad optimization we put into this product, to give you the cleanest and most productive experience.