Muapi Changelog

Lyria 2 is Google's instrumental music generation model. Describe the style and mood in a text prompt and get a roughly 30-second instrumental track as an MP3.

Lyria 3 generates short, high-quality music clips of up to 30 seconds from a text prompt — perfect for jingles, intros, loops and background tracks. Returned as an MP3.

Lyria 3 Pro is Google's flagship music model: describe a genre, mood, instruments and structure and it composes a full-length song of up to three minutes, returned as an MP3.

Fish Audio S2.1 Pro is a high-fidelity text-to-speech model built for long-form narration. Send up to 10,000 characters in one request and get back a natural-sounding MP3. Billed per UTF-8 byte of text.

Eleven v4 is ElevenLabs' latest text-to-speech model, delivering highly expressive, natural single-voice narration. Send up to 2,000 characters and get back a studio-quality MP3 — ideal for voiceovers, audiobooks, ads and app audio.

Ideogram 4.0 is Ideogram's latest text-to-image model, built for sharp, legible typography, clean graphic design and photorealistic scenes. Generate posters, logos, packaging and text-heavy visuals at 1K or 2K in five aspect ratios from prompts up to 4,000 characters.

Gemini 4 Argon is Google DeepMind's frontier reasoning model with a 1M-token unified context and output window, built for long-horizon coding, knowledge work, and cybersecurity. Accepts text and image input with an optional system prompt. Token-based pricing: $2.00/M input tokens and $10.00/M output tokens.

Vidu Q4 Reference-to-Video builds a 3-16 second clip at up to 4K from up to 15 reference images and 3 reference audio clips, keeping subjects consistent and adding synchronized dialogue and sound effects.

Vidu Q4 Image-to-Video animates a first-frame image into a 3-16 second clip at up to 4K with synchronized dialogue and sound effects. It inherits the aspect ratio of the source image and follows an optional motion prompt.

Claude Haiku 5.5 is Anthropic's fastest and most cost-effective model, built for high-frequency queries, simple tasks, and near-instant response times. Supports text and image inputs. Token-based pricing: $0.60/M input tokens, $3.00/M output tokens. Two endpoints: standard async (/claude-haiku-5-5) and live streaming (/claude-haiku-5-5/stream) via SSE.

Nano Banana 2.1 is Google's latest Gemini image model, delivering fast, high-fidelity image generation and editing up to 4K with strong prompt adherence and legible in-image text.

Nano Banana 2.1 is Google's latest Gemini image model, delivering fast, high-fidelity image generation up to 4K with strong prompt adherence and legible in-image text.

GPT 6.1 Sol is OpenAI's GPT-6 model positioned between Astra and Luna, delivering near-Astra agentic coding, computer-use, and professional-work capability at one-fifth of Astra's token price. Supports mixed text and image inputs, adjustable reasoning effort, and integrated web search. Pricing: $2.00/M input tokens, $10.00/M output tokens.

Claude Sonnet 5.5 is Anthropic's faster, more cost-efficient upgrade to Sonnet 5, built for everyday coding, agentic workflows, knowledge work, and visual understanding. Supports text and image inputs. Token-based pricing: $2.00/M input tokens, $10.00/M output tokens. Two endpoints: standard async (/claude-sonnet-5-5) and live streaming (/claude-sonnet-5-5/stream) via SSE.

Decompose a composite image into a base image plus clean, separate layer assets with Seedream 5.0 Flash Layer Decomposition. Optionally target elements with prompts or <bbox> regions.

Seedream 5.0 Flash Edit is ByteDance's fast, low-cost image-to-image tier, editing or combining up to 10 reference images from a text instruction at 1K or 2K.

Seedream 5.0 Flash is ByteDance's fast, low-cost Seedream 5 text-to-image tier, delivering photorealistic 1K and 2K images with strong prompt adherence and in-image text for high-volume creative workflows.

Hermes 4 405B is a 405B hybrid-reasoning chat model from Nous Research with a 128K-token context window, known for steerability and low refusal rates.

Venice Abliterated is a general-purpose chat model with a 128K-token context window that is tuned to refuse less often than mainstream assistants.

Gemma 4 31B SDFT Abliterated is a Gemma 4 31B variant with a 262K-token context window and image input, modified to reduce refusal behavior and tuned for creative chat; see also the [Character Chat LLM API](/character-chat-llm-api).

Qwen3.5 27B Opus-Distilled Derestricted is a derestricted Qwen3.5 27B reasoning-distilled model with a 262K-token context window and image input, modified to reduce refusal behavior.

Qwen3.5 27B Blossom Derestricted is a derestricted Qwen3.5 27B model with a 262K-token context window and image input, modified to reduce refusal behavior.

Qwen3.5 27B Queen Derestricted is a derestricted Qwen3.5 27B model with a 262K-token context window and image input, modified to reduce refusal behavior.

GLM 4.6 Derestricted v5 is a derestricted GLM 4.6 reasoning model with a 131K-token context window, modified to reduce refusal behavior.

Qwen 3.8 27B Fable is a 27B Qwen 3.8-based community fine-tune tuned for creative writing, character chat and interactive fiction (it is not abliterated) geared toward long-form storytelling, with a 524K-token context window and image input.

Qwen 3.8 27B Queen is a 27B Qwen 3.8-based community fine-tune tuned for creative writing, character chat and interactive fiction (it is not abliterated), with a 524K-token context window and image input.

Aion Llama 3.1 8B is an 8B Llama 3.1-based community fine-tune tuned for creative writing, character chat and interactive fiction (it is not abliterated), with a 32K-token context window and tool-calling support.

Nevoria R1 70B is a 70B Llama 3.3-based merged community fine-tune tuned for creative writing, character chat and interactive fiction (it is not abliterated), with a 32K-token context window.

Wayfarer 70B is a 70B Llama 3.3-based community fine-tune tuned for creative writing, character chat and interactive fiction (it is not abliterated) built for adventure and text-game style storytelling, with a 32K-token context window.

MythoMax 13B is a classic 13B Llama 2-based community fine-tune tuned for creative writing, character chat and interactive fiction (it is not abliterated), with a short 4K-token context window.

EVA Llama 3.33 70B is a 70B Llama 3.3-based community fine-tune tuned for creative writing, character chat and interactive fiction (it is not abliterated), with a 32K-token context window.

Magnum v4 72B is a 72B community fine-tune tuned for creative writing, character chat and interactive fiction (it is not abliterated) aimed at prose quality, with a 16K-token context window.

Stheno 8B v3.2 is an 8B Llama 3-based community fine-tune tuned for creative writing, character chat and interactive fiction (it is not abliterated), with a 16K-token context window.

Euryale 70B v2.3 is a 70B Llama 3.3-based community fine-tune tuned for creative writing, character chat and interactive fiction (it is not abliterated), with a 20K-token context window.

Anubis 70B v1.1 is a 70B Llama-based community fine-tune tuned for creative writing, character chat and interactive fiction (it is not abliterated), with a 32K-token context window.

Cydonia 24B v4.1 is a 24B community fine-tune tuned for creative writing, character chat and interactive fiction (it is not abliterated), with a 131K-token context window.

Gemma 4 31B Gembrain Abliterated is a Gemma 4 31B community variant with a 262K-token context window, image input and tool-calling support, modified to reduce refusal behavior.

GLM 5.3 Abliterated is a full-size GLM 5.3 reasoning model with a 1M-token context window and tool-calling support, modified with abliteration to reduce refusal behavior.

Gemma 4 26B A4B Abliterated is a Gemma 4 26B A4B mixture-of-experts model with a 65K-token context window and tool-calling support, modified with abliteration to reduce refusal behavior.

GLM 5.3 Flash Abliterated is a fast GLM 5.3 Flash reasoning model with a 1M-token context window, image input and tool-calling support, modified with abliteration to reduce refusal behavior.

Qwen 3.8 27B Abliterated is a 27B Qwen 3.8 model with a 524K-token context window and image input, modified with abliteration to reduce refusal behavior; an optional thinking mode enables extended reasoning.

NeuralDaredevil 8B Abliterated is a lightweight 8B abliterated chat model with an 8K-token context window, suited to low-cost prompts.

Qwen 2.5 32B Abliterated is an abliterated version of Qwen 2.5 32B Instruct with a 32K-token context window, modified to reduce refusal behavior.

DeepSeek R1 Distill Qwen 32B Abliterated is a compact abliterated DeepSeek R1 reasoning distillation built on Qwen 32B with a 16K-token context window.

DeepSeek R1 Distill Llama 70B Abliterated is an abliterated DeepSeek R1 reasoning distillation built on Llama 70B with a 16K-token context window.

Llama 3.3 70B Abliterated is an abliterated version of Meta's Llama 3.3 70B Instruct with a 32K-token context window, modified to reduce refusal behavior.

Qwen 3.8 27B Obliterated is a 27B Qwen 3.8 model with a 524K-token context window and image input, modified to reduce refusal behavior; an optional thinking mode enables extended reasoning.

Abliterated Model Large V2 is the second revision of the large abliterated reasoning model, with a 1M-token context window and tool-calling support.

Abliterated Model Large is a large text reasoning model with a 1M-token context window and tool-calling support, modified with abliteration to reduce refusal behavior.

Abliterated Model is a multimodal reasoning model with a 262K-token context window and image input, modified with abliteration to reduce refusal behavior.

MiMo V2.6 Flash Abliterated is a fast reasoning model from the MiMo V2.6 Flash family with a 1M-token context window, image input, and tool-calling support, modified with abliteration to reduce refusal behavior.

Claude Opus 5.5 is Anthropic's next-generation Opus model, delivering top-tier coding, long-context reasoning, and agentic capabilities at a lower price than Opus 5. Supports text and image inputs. Token-based pricing: $4.00/M input tokens, $20.00/M output tokens. Two endpoints: standard async (/claude-opus-5-5) and live streaming (/claude-opus-5-5/stream) via SSE.

Gemini 3.8 Flash Lite TTS is Google's high-throughput, low-latency, cost-efficient text-to-speech model for expressive multi-speaker conversational audio. Control voice, accent, emotional style, and pace per speaker for high-volume voiceovers, real-time dialogue, and budget-conscious narration.

Gemini 3.8 Flash TTS turns written dialogue into expressive, natural multi-speaker speech with studio-grade voice fidelity, expressive acting, and long-form stability. Control voice, accent, emotional style, and pace per speaker for fast, affordable voiceovers, character dialogue, and narration.

Fundamental financial metrics/ratios (historical or latest snapshot) for a ticker, or a filter-based screener across US public companies. Read-only research data only — no trading, portfolio, or investment-advice output.

Daily end-of-day stock price history (OHLCV bars) by ticker, over a date range. Read-only research data only — no trading, portfolio, or investment-advice output.

Public crypto market data: live price quote, coin profile, historical price series, top-coins market overview, or coin search. Read-only research data only — no wallet, trading, portfolio, or investment-advice output.

Public-company research data by stock ticker: profile, standardized financial statements, SEC filing metadata, or a real-time quote. Read-only research data only — no trading, portfolio, or investment-advice output.

Amazon-native keyword search-volume metrics, distinct from Google/web keyword volume.

Retrieve public reviews for a marketplace product, where the upstream source supports review text/pagination. Currently supported for tiktok_shop only.