Models

Pass provider and model on any generation request. Omit model to use that provider's default. The authoritative, always-current catalogue is served by GET /v1/providers — treat this page as reference.

GET /v1/providers

curl https://api.genzoai.com/v1/providers?type=text \
  -H "Authorization: Bearer $GENZO_API_KEY"

Text

Model	Provider	Context	Cost	Status	Best for
claude-haiku-4-5	Anthropic	200K	$	Live	Classification, extraction, high-volume simple work
claude-sonnet-5	Anthropic	1M	$$	Live	The default choice — near-flagship quality at moderate cost
claude-opus-5	Anthropic	1M	$$$	Live	Hard reasoning, long agentic runs, complex code
gpt-5	OpenAI	—	$$	Live	General purpose, strong tool use
gpt-5-mini	OpenAI	—	$	Live	Cheap, fast, good enough for routing and short tasks
gemini-3.6-flash	Google	1M	$$	Live	Long documents, fast multimodal reasoning
gemini-3.5-flash	Google	1M	$	Live	High-throughput work at low cost

Coming means the provider is registered but not yet callable — requests naming it return 503. Everything marked Live works today with any active key.

Reference list prices

What the upstream provider charges, per million tokens. Your credit cost derives from these plus the GenzoAI margin — call GET /v1/credits/estimate for the exact figure before you commit to a model.

Model	Input / 1M	Output / 1M	Relative
claude-haiku-4-5	$1.00	$5.00	1×
claude-sonnet-5	$3.00	$15.00	3×
claude-opus-5	$5.00	$25.00	5×

What you are charged today. Credits are a flat cost per modality and provider, not per token — so the list prices below are the upstream vendor's rates for reference, not your bill. Call GET /v1/credits/estimate for the figure that will actually be deducted. Per-model, per-token pricing is on the roadmap.

Image

Model	Provider	Sizes	Billing	Status
gpt-image-1	OpenAI	256² – 1792×1024	Per image, ×2 at `hd`	Live

Video

Model	Provider	Billing	Latency	Status
gen4.5	Runway	Per second of output	1–4 min	Live
gen4_turbo	Runway	Per second of output	< 1 min	Live

gen4_turbo is image-to-video only — it requires a source image.

Speech

Model	Provider	Billing	Status
tts-1	VoiceNom	Per 1,000 input characters	Live

Choosing a model

Start at Sonnet. It handles most production work; move up only where your evals show you need it.
Drop to Haiku for volume. Classification, routing, extraction and summarisation rarely justify a flagship model.
Reserve Opus for hard reasoning. Multi-step analysis, long agentic runs, difficult code.
Benchmark before committing. Only the model field changes — run the same prompt across three models and compare cost against quality on your own data.

Deprecation policy

Models are removed only after the upstream provider retires them. We give at least 30 days' notice by email to every account that called the model in the previous 30 days, and publish a replacement recommendation. Requests naming a retired model return 404 with the suggested successor in the error body.