**What’s New in 1.2026.05** - Improved iOS chat reliability when using web search and image attachments. - Fixed cases where local MLX models could stop before generating a response. - Added clearer fallback messages when a generation stops unexpectedly. - Improved streaming stability so completed responses are preserved more reliably. - Reduced jarring scroll movement when sending messages in existing conversations. - Restored Ministral 3 model availability in the model catalog. - Added an in-app What’s New screen for release highlights. - Updated the app version shown in Settings to `1.2026.05`.

What’s New * Run code directly with Artifacts for a more capable workspace. * Added Real-Time Web Search, powered by Oscilla’s private search engine. * Revamped Reasoning so you can better follow how supported models think through complex prompts. * Added Reasoning for Gemma 4 (E2B) * Added new local models with improved options for speed, intelligence, and on-device use. * Nvidia Nemotron Mini 4B * IBM Granite 4.1 (3B & 8B) * LFM 2.5 (350M, 1.2B Instruct, 450M VL) * OpenBMB MiniCPM-V 4.6 * Improved onboarding and model selection for devices that need local model downloads. * Improved web citations so sources are cleaner and easier to open. * Improved Streaming Run Time * Fixed Storage Bug * Fixed Bugs * Improved UX/UI

What’s New * Run code directly with Artifacts for a more capable workspace. * Added Real-Time Web Search, powered by Oscilla’s private search engine. * Revamped Reasoning so you can better follow how supported models think through complex prompts. * Added Reasoning for Gemma 4 (E2B) * Added new local models with improved options for speed, intelligence, and on-device use. * Nvidia Nemotron Mini 4B * IBM Granite 4.1 (3B & 8B) * LFM 2.5 (350M, 1.2B Instruct, 450M VL) * OpenBMB MiniCPM-V 4.6 * Improved onboarding and model selection for devices that need local model downloads. * Improved web citations so sources are cleaner and easier to open. * Improved Streaming Run Time * Fixed Storage Bug * Fixed Bugs * Improved UX/UI

This update improves local reasoning models, vision handling, and the chat experience across Oscilla. - Fixed reasoning support for Qwen 3 and Qwen 3.5, including cleaner thinking controls, better parsing, and safer defaults. - Added proper thinking support for SmolLM3 and LFM 2.5 Thinking. - Added Gemma 4 reasoning support with model-specific thinking behavior. - Reasoning is now opt-in by default for supported models, and always disabled in Voice Mode. - Removed reasoning and Voice Mode controls from LFM 2.5 Thinking where they do not apply. - Improved the reasoning UI with a cleaner Working, Thinking, and Analyzing state. - Added new reasoning UX/UI. - Improved image routing for Qwen 3.5 so text and vision paths are handled more safely. - Added safeguards for Qwen 3.5 vision on lower-memory devices. - Improved chat scroll behavior when sending messages and rerunning responses.