Mr Broccoli 4.2 brings the newest AI models, better speech, and a long list of reliability fixes. NEW MODELS • GPT-6 Astra, GPT-6 Sol and GPT-6 Luna, with every supported reasoning effort up to Max. • Claude Opus 5.5, Claude Opus 5 and Claude Fable 5.1. • Grok 4.7 and Grok 4.6. • Gemini 3.8 Flash and Gemini 3.7 Flash. • Qwen 3.8 Max, Qwen 3.8 Flash and Qwen 3.7 Flash, with model-specific reasoning controls. • DeepSeek V4.1 Flash, which now also understands images. DeepSeek models add a Low reasoning setting. • OpenRouter adds current versions of GPT-6 Astra, Sol and Luna, Claude Opus 5.5, Opus 5 and Fable 5.1, Grok 4.7 and 4.6, Gemini 3.8 Flash, Qwen 3.8 Max, and DeepSeek V4 Pro and V4.1 Flash. SPEECH • GPT Transcribe is the new default for OpenAI speech input, and Gemini 3.5 Transcribe adds word-for-word transcription. • New Gemini 3.8 Flash and Flash-Lite voices for spoken replies. • Long dictation now keeps every sentence, not just the last one. • System Recognition can be chosen whenever iOS offers it, even when strict offline recognition is unavailable. • Full reply first pauses between paragraphs again, and cancelling reliably stops any queued audio. • Voice previews can now replace a paused spoken reply. • Voices saved for a conversation stay selected when a provider updates its models, and OpenAI's older voice models only show voices they support. WORKSPACE • A cleaner voice and text layout: the arrows, voice orb and composer sit on one line, Image and Hands free move to the edges, and Council and Web stay centred. The composer and send button stay visible on small screens and with the keyboard open. • Adding an answering model no longer leaves Settings unresponsive. RELIABILITY AND PRIVACY • Starting a second recording or Hands free no longer stops a recording already in progress. • Web search keeps usable results, rejects empty answers and cancels stalled responses. Gemini and xAI search no longer ask the provider to store your requests. • Model Council now credits the provider and model that actually answered. • Backup restores and conversation branches report failed saves instead of claiming success. • Locked conversations stay locked if the app goes to the background while one is loading. • The app recovers from temporary storage start-up problems without a restart. • A damaged settings file no longer hides your saved API keys. • Slow requests always end with a clear timeout message. CHANGES • Qwen speech input and output are removed ahead of Alibaba retiring them on October 10. Qwen chat and search are unchanged, and saved Qwen voices switch to your device's built-in speech. • Google now limits Gemini 2.5 to accounts that already used it, so it is no longer offered for new choices. Existing selections keep working, and older Gemini 2.5 voices move to Gemini 3.8 with the same voice. • OpenAI Realtime models are hidden until they are fully supported. Saved selections move to GPT-5.6 Sol or GPT-5.4 mini. • The ElevenLabs built-in fallback voice is now correctly named Janet.

Mr Broccoli 4.1 focuses fully on BYOK: the paid app opens straight into the workspace, without Free/local-response mode or onboarding; optional local speech remains. Add unlimited response routes, use the redesigned voice controls and Hands free mode, and follow live Model Council progress. Transcript, images, local speech and encrypted backups are improved, with fixes for long Council readback and many interactions.