Text Generator Blog - Text Generation Research

Five-Minute Training Runs: How the Roleplay Model Trains Now
Our roleplay QLoRA serves at 264 tok/s, and every future run is a 5-minute capped, resumable micro-run on rented GPUs with a budget ledger

Inverse Scaling Prize - 250k in prizes for research
The inverse scaling prize for finding hard datasets that larger models do worse on

New Unfiltered Chat GPT Model for Roleplay and Chat
A new transformer model trained on unfiltered chat data from WizardLM and Pygmalion datasets

Prompt Tuning - How to make realistic text
Generating realistic text is sensitive to formatting. Learn prompt tuning tips for consistent, natural output from language models

10x OpenAI Cost Savings - One Line Change
Change this one line to save over 10x on the cost of AI Text+Code Generation

Many Shots Are Better Than One for Language Models
The more examples, the better your model will generalize, Language models can understand repeating patterns with more shots

Text Generator Now Researches via Crawling
Crawling Text and Images Aids Text Generator to Generate Text From Unseen Images and Links

Embed Images, Text, and Code in the Same Space
Embed Images, Text, and Code for analysing multiple content types together

Cost Effective Speech to Text API
Speech to Text API over 8x cheaper than Google Cloud. See how to transcribe audio with a simple, low cost API call

Text Generator May Go Down Due to Cyclone Gabrielle
Text Generator may experience downtime due to Cyclone Gabrielle in New Zealand. Read the status update and expected recovery

Generate Speech with Text-Generator.io
Text Generator can understand speech, images, code, it can reply and now can talk too

Summarization API Now Accepts Max Length Parameter
Control summary length precisely by specifying maximum characters in the summarization API

New Thinking Model - Qwen3.5 with Reasoning Support
Text Generator upgraded to Qwen3.5-4B with built-in thinking and reasoning, OpenAI-compatible chat completions API with streaming

Gemma 4 Multimodal Audio API - Audio In, Text + Audio Out
New API endpoint powered by Gemma 4 and Kokoro TTS: send audio and text, get intelligent text responses with natural speech audio in a single request

Kimi K3 API - Moonshot's 2.8T Open Model with 1M Context
Kimi K3 is now on the Text Generator chat completions API: 2.8T parameters, 1M token context, vision and tool calling at $3/$15 per million tokens

Choosing the Right LLM for Each Task
A practical guide to matching model size, context window, and price to the job, with the Text Generator auto profiles that do the matching for you

LLM Routing: One API Key for Many Providers
How a single OpenAI compatible endpoint can reach OpenAI, Anthropic, Google, xAI, DeepSeek, Moonshot and more, with fallbacks and one bill

Text, Vision and Speech APIs in One Place
Use Text Generator for chat, image understanding, embeddings, speech to text, and text to speech through OpenAI compatible endpoints

Blog Posts