Model catalog

Directorial TTS with bracketed emotional cues. 100 chars/sec.

Expressive TTS with emotional control. Sighs, whispers, cues.

Native bidirectional audio streaming. Sub-200ms latency.

228x speed transcription on Groq. Ultra-low cost STT.

Large MoE (22B active) for complex multilingual tasks.

Dual-mode thinking/non-thinking. 662 TPS on Groq hardware.

Ultra-fast responses at low cost. Ideal for high-throughput.

Most powerful Claude. Extended thinking, 1M beta context window.

Balanced performance and speed for enterprise workloads.

671B MoE (37B active). Extreme efficiency with sparse attention.

Fastest Gemini variant at near-zero cost.

2M context with native Google Search grounding.

Latest Gemini with advanced vibe coding and multimodality.

400B MoE (17B active). Native multimodal, 562 TPS on Groq.

109B MoE (17B active). Lean multimodal, near 600 TPS.

Enterprise-grade with 256K context. EU data sovereignty.

Efficient model for routine tasks and high volume.

1T MoE (32B active). Excels at frontend dev and tool calling.

Reliable general-purpose model with function calling and vision.

Cost-efficient model for high-volume production tasks.

Ultra-low cost for triage, extraction, and metadata tasks.

Latest OpenAI flagship. 400K context with prompt caching support.

Open-weight MoE (5.1B active). Optimized for agentic workflows.

Compact 20B model. Over 1,000 TPS on Groq LPU hardware.

2M context with fast reasoning and competitive output pricing.

Massive 480B coding MoE (35B active). Top benchmark scores.

Specialized for code generation, completion, and refactoring.

Open model tailored for code agents and automation.

Optimized for fast code generation and debugging.

1024-dim English embeddings. Native image embedding support.

Task-specific LoRA adapters. 1024 dims, truncatable to 32.

3072-dimensional high-accuracy embeddings.

Cost-efficient 1536-dimensional embeddings.

First MoE embedding model. 1024 dims, Matryoshka support.

Cost-optimized embeddings with flexible dimensions.

Professional quality generation with high detail.

Ultra-fast 4-step generation at minimal cost.

Premium high-fidelity synthesis. 50 diffusion steps.

Best-in-class text rendering accuracy in generated images.

Text-to-image and multi-image style transfers.

Tokenized image generation. DALL-E successor with precise prompting.

Fast distilled generation. 2.5 credits per generation.

High-fidelity diffusion model. 6.5 credits per generation.

Balanced quality and speed. 3.5 credits per generation.

Reasoning specialist with 23K internal thinking tokens. AIME SOTA.

Advanced reasoning for complex multi-step deduction.

Premium reasoning with highest accuracy. Math and crypto focus.

Cost-effective reasoning model for everyday logic tasks.

Advanced video generation architecture via Runway API.

Cinematic realism with complex camera motions. 1080p.

State-of-the-art physics simulation. 720p with synced audio.

Premium 1080p video generation with native audio.

Fast video generation. 5 credits per second.