Newsroom
New models and toolsets as we ship them — what changed and how to use it, in chat, via API or over MCP.
Google music generation in Twelver now runs on Lyria 3.5, the stable release that replaces the Lyria 3 Pro preview. A full song of a couple of minutes with verses, choruses and vocals in the language of your prompt, at the same price. Lyria Clip stays for 30-second drafts. Available in Twelver chat and over MCP.
Google speech in Twelver now runs on Gemini 3.8 Flash TTS, the stable release that replaces the preview. Two-speaker dialogue in one file, delivery described in plain words, and dozens of languages. A second of speech costs about 10% less. Available in Twelver chat and over MCP.
Alibaba announced Qwen 4 at the Apsara Conference on 22 September 2026: four models — Max, Plus, Flash and 27B — on a new architecture. There is no release date yet. We’ll add Qwen 4 to Twelver as soon as it ships; Qwen3.8 is live in Twelver today.
DeepSeek has confirmed V4.1 Pro, the flagship of its new V4.1 generation. The smaller V4.1 Flash shipped on 10 September 2026; Pro has no date yet. Meanwhile DeepSeek V4 Pro and V4 Flash are live in Twelver.
Grok video in Twelver is now made by Grok Imagine Video 1.5: there’s a new 1080p tier, and an animated photo holds on to the face and detail of the original much better. Clips run 1 to 15 seconds, with sound. Available in Twelver chat and over MCP.
Grok images in Twelver are now drawn by Grok Imagine Image 2.0 — frames look like photographs rather than illustrations. Text-to-image, editing a shot and building a frame from several photos work as before. Available in Twelver chat and over MCP.
Wan is now 3.0: one model shoots from text, animates a frame, builds a clip from references, and edits or extends existing footage. Clips run up to 30 seconds — twice as long as before — with dialogue, sound and music. Available in Twelver chat and over MCP.
GPT-6 Luna replaces GPT-5.6 Luna: newer, with fresher knowledge and half the input price. The context is unchanged at 1.05M tokens. Available over the OpenAI-compatible API under the same gpt name.
Grok 4.6 is replaced by Grok 4.7 at the same rate: a 500K-token context, text and image input, and reasoning effort now up to xhigh. Available over the OpenAI-compatible API.
Our hosting provider had an incident, and some recent data was lost. We moved Twelver to new servers, restored the database from a backup, and are bringing back what can be recovered: payments, streaks and files.
GLM 5.2 is replaced by GLM 5.3 at identical rates, and GLM 5.3 Flash joins it — natively multimodal with a 1.3M-token context. 5.3 is what chat routes to; both are callable over the OpenAI-compatible API.
Qwen3.8 Max replaces 3.7 Max and costs less than it did; Qwen3.8 Flash has the cheapest input of any model we serve. Both carry a 1M-token context. Available over the OpenAI-compatible API.
DeepSeek V4 Pro joins V4 Flash — a 1.6T-parameter flagship. Both carry a native 1M-token context and a thinking mode that can genuinely be switched off. Available over the OpenAI-compatible API.
Grok 4.5 is replaced by Grok 4.6 at the same rate: a 500K-token context, text and image input, selectable reasoning effort. Available over the OpenAI-compatible API.
Seedance 2.0 is a new video-generation model: text-to-video, photo animation and more coherent short scenes. Available now in Twelver chat and over MCP for third-party clients (Claude, Cursor).
GLM 5.2 is a new text model: strong reasoning, long context and coding. Available in Twelver chat and via the OpenAI-compatible API for agents like Cline, OpenCode and Cursor.
Seedream is an image-generation model: text-to-image, editing and blending multiple references. Available in Twelver chat and over MCP for Claude, Cursor and other clients.
Nano Banana is Google’s image generation and editing model: text-to-image, edit-by-description and faithful preservation of faces and objects. Available in Twelver chat and over MCP for Claude, Cursor and other clients.
Grok now generates both images and video — from text or an uploaded frame. Available in Twelver chat and over MCP for Claude, Cursor and other clients.
Veo 3.1 is Google’s video-generation model: text-to-video, photo animation and scene extension. Available in Twelver chat and over MCP for Claude, Cursor and other clients.
Hailuo is MiniMax’s video-generation model: text-to-video, photo animation and scenes with a given subject. Available in Twelver chat and over MCP for Claude, Cursor and other clients.
We’ve updated HappyHorse to 1.1 — a video-generation model: text-to-video, photo animation and reworking an existing clip. Available in Twelver chat and over MCP for Claude, Cursor and other clients.
Lyria 3 is Google’s music-generation model: a track from a text description in any genre and mood. Available in Twelver chat and over MCP for Claude, Cursor and other clients.
MiniMax Music is a music-generation model: a track from a description and a cover of an existing melody. Available in Twelver chat and over MCP for Claude, Cursor and other clients.
Gemini TTS is Google’s speech model: natural voiceovers, multi-voice dialogue and voice design. Available in Twelver chat and over MCP for Claude, Cursor and other clients.
Qwen3 TTS is a speech model: natural voiceovers, voice cloning and voice design. Available in Twelver chat and over MCP for Claude, Cursor and other clients.