gpt-oss-20b delivers strong performance with fast inference and efficient reasoning, and is available on every plan, including Free.

Mistral AI's mid-size model, balancing capability and cost with a 128K context window.

A 27B-parameter multimodal model from Alibaba's Qwen team, built as an image-text-to-text model with a 256K-token context window and function calling.

Meta's Llama 3.3 — a 70B instruction-tuned model and one of the most widely benchmarked open models available, with function calling support.

A cutting-edge model designed for high-level reasoning, instruction-following, and advanced agent capabilities — strong across both general-purpose and specialized applications.

Alibaba's vision-language model, built specifically to understand and reason about images alongside text.

Alibaba's largest Qwen3.5 model — 397B parameters and a 256K context window, built for the most demanding reasoning and multimodal tasks.

An LLM-based multilingual embedding model trained across many languages (English, Chinese, Japanese, Korean, French, and more) and task types, with state-of-the-art results on multilingual benchmarks like MIRACL and MTEB.

Alibaba's Qwen3-Embedding-4B — built on the Qwen3 dense model family for text embedding and ranking, with strong multilingual, retrieval, and code-retrieval performance. Supports 100+ languages and user-configurable output dimensions (32 to 2560). Apache 2.0.

Stability AI's SDXL — a widely-used open image generation model, offered on this platform at no cost.

OpenAI's Whisper — automatic speech recognition and translation trained on over 5M hours of labeled audio, with strong generalization across accents, noise, and domains. Supports 99 languages. Apache 2.0.

Google DeepMind's Gemma 4 26B A4B — an instruction-tuned mixture-of-experts model with 25.2B total parameters but only 3.8B active per token, giving near-31B quality at MoE-efficient cost. Genuinely multimodal (text, images, and short video), with a 256K context window, native function calling, a toggleable reasoning mode, and structured output support. Scores 77.1% on LiveCodeBench v6 and 88.3% on AIME 2026. Apache 2.0. For vision requests, set max_tokens to at least 1000–1500 — the model spends part of its budget on internal reasoning before producing a visible answer.

A compact 0.9B-parameter vision-language model purpose-built for document parsing, pairing a dynamic-resolution visual encoder with a lightweight ERNIE-4.5 language model. We run version 1.6, which scores 96.33% on OmniDocBench v1.6 (state-of-the-art) and adds seal/stamp recognition, text spotting, and improved Chinese ancient-document and rare-character handling on top of the base model's tables, formulas, and charts across 109 languages. Apache 2.0.

A pruned, finetuned version of Whisper Large v3 — same transcription model with decoding layers reduced from 32 to 4, trading a little accuracy for meaningfully faster turnaround. Supports 99 languages. MIT licensed.

Kokoro is a lightweight, open-weight TTS model — just 82 million parameters on a decoder-only StyleTTS2 + ISTFTNet architecture, no diffusion. Despite its size it delivers quality comparable to much larger TTS models while running significantly faster and cheaper, with 54 voices across 8 languages. Trained exclusively on permissive, non-copyrighted audio. Apache 2.0.

Zhipu AI's GLM 5.2.

DeepSeek's fast, low-cost V4.1 variant.

DeepSeek's flagship V4 Pro.

Moonshot AI's Kimi K2.7, tuned for code.

Moonshot AI's flagship Kimi K3.

MiniMax's M3.