Models

Free experimental stealth model available on Siron.dev while in preview. Rate-limited, no cost.

Cheap, quick multimodal model with a 1M context window. Currently discounted for launch.

The workhorse Claude: near-Opus quality at a fraction of the cost. Excellent for production agents.

Open-weight workhorse with a huge context window at a friendly price. Solid general assistant baseline.

Balanced quality and cost for everyday assistant and agent workloads. Strong instruction following at Mini pricing.

Fastest Claude tier for classification, extraction and lightweight chat at scale.

Ultra-fast general chat model tuned for high throughput and low latency. Great default for assistants and high-volume pipelines.

Massive 1M-token context with native image, audio and document understanding. Great for whole-repo or long-video analysis.

Larger Llama 4 tier for reasoning-heavy open-weight deployments and long document synthesis.

Flagship general model with frontier reasoning, tool use and long-horizon agentic reliability.

Agentic open-weight model with very long context and strong tool-use benchmarks at low cost.

Chain-of-thought reasoning model for math, logic and hard multi-step problems. Emits explicit reasoning traces.

Open-weight MoE instruct model with strong multilingual coverage, especially CJK and Southeast Asian languages.

Frontier reasoning and coding with long-horizon agentic reliability. Best-in-class at multi-file refactors.

Specialized coding model with strong repo-level and agentic performance across 90+ languages.

Community-tuned creative writing and roleplay model with a distinctive voice and long memory.

European flagship with excellent function calling, JSON mode and low-latency streaming.

Steerable open-weight assistant with minimal refusals, strong persona control and clean tool calling.

Vision-enabled variant of V4 Flash. Reads screenshots, charts and documents while keeping Flash-tier latency.

Real-time-aware model with aggressive reasoning and a large context. Strong at current-events grounded answers.

Speech-to-text transcription and translation across 100+ languages, billed per minute of audio.

State-of-the-art text-to-image generation with strong prompt adherence and typography. Billed per output image.

Deliberate long-thinking model for research-grade math, proofs and complex planning. Higher latency by design.

Photorealistic image generation with fine control over lighting, lens and composition.

Text-and-image-to-video generation with synchronized audio, up to 20 second clips.