Choosing a Model

Match models to your task, latency, and cost targets. All models are called through one base URL:

https://api.sagea.space/v1

Quick Selection Guide

Generalist models (text, reasoning, coding, agents)

Model	Best for	Context	License
SAGE 2.4 Actus	Frontier agentic + coding, multimodal	128k	Modified MIT
SAGE 2.5 Celer	Fast instruct/reasoning/coding in one model	32k	Apache 2.0
SAGE OSS	Self-host, fine-tune, research (7B/14B/24B)	16k	Apache 2.0

Specialized models (audio, OCR)

Model	Best for	Pricing
Sonus TTS	TTS + zero-shot voice cloning, 30+ languages	$15 / 1M chars standard
ARVA OCR	Document AI, tables, handwriting, 150+ languages	$1.00 / 1,000 pages

Decision Framework

1. Define Your Use Case

Complex reasoning / agents / coding: Choose SAGE 2.4 Actus. 72B, 128k context, function calling, structured outputs, /v1/agents.

curl -X POST https://api.sagea.space/v1/chat/completions \
  -H "Authorization: Bearer $SAGEA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "sage-2-4-actus", "messages": [{"role": "user", "content": "Analyze the implications of quantum computing on cryptography"}]}'

High-throughput / cost-sensitive: Choose SAGE 2.5 Celer. 24B, 32k context, 3x faster at equivalent quality. $0.10 / $0.30 per 1M tokens.

Self-hosted / fine-tuning: Choose SAGE OSS. Open weights, LoRA + full fine-tune support.

Voice: Choose Sonus TTS for cloning + streaming at /v1/audio/speech. See Voice quickstart.

Documents: Choose ARVA OCR for layout-aware extraction at /v1/ocr/process. See Document quickstart.

2. Compare performance

Requirement	Actus 2.4	Celer 2.5	OSS	Sonus TTS	ARVA OCR
Quality	Highest	High	Good	Studio	99.2% OCR
Speed	Fast	Fastest	Fastest	`sub-200ms` first byte	Fast batch
Context	128k	32k	16k	4k chars/input	100MB/file
Cost	$0.40/$1.20	$0.10/$0.30	$0.05/$0.15	$15/1M chars	$1.00/1K pages

Per-model rates are listed on each model card.

3. Evaluate Technical Constraints

Cloud-only: any model works.
Edge/mobile/offline: Celer or OSS for text; Sonus streaming for audio.
Real-time: Celer (chat) + Sonus stream (audio).
High throughput: Celer + batch at /v1/batch.

Cost Optimization

Cache generated audio and repeated prompts.
Use the right model: Celer for simple Q&A, Actus for reasoning.
Batch processing: /v1/batch for thousands of requests.
Right context: don't send 128k when 32k suffices.

Migration Strategies

Test thoroughly: compare quality and latency in Foundry.
Gradual rollout: A/B Actus vs Celer.
Monitor metrics: latency, quality, cost per 1K requests.
Fallback plan: keep previous model string as backup.