Model Catalog & Pricing
The AI models Edgaze workflows can call, what hosted runs cost per model, and how older model ids are automatically migrated to current ones.
Overview#
Every LLM block in an Edgaze workflow (LLM Chat, LLM Image, Embeddings, AI Conditions) runs against a curated model catalog. This page lists the current catalog, the hosted platform rate for each model, and explains what happens to workflows that reference older models.
Rates below are the hosted platform rates used for run billing, cost estimation, and holds. They track provider list pricing and include a small platform margin that covers orchestration, retries, and infrastructure variance. This page renders live from the same model catalog the builder and billing engine use, so it can never fall out of date.
For how per-run estimates are produced, read Infrastructure Cost Estimation. For how estimates combine with creator margin, read Marketplace Fees.
Chat Models#
| Model | Best for | Input (per 1M tokens) | Output (per 1M tokens) |
|---|---|---|---|
| GPT-5.6 TerraRecommended | High-quality general workflows | $2.70 | $16.20 |
| GPT-5.6 Sol | Flagship reasoning | $5.40 | $32.40 |
| GPT-5.6 Luna | Fast, high-volume steps | $1.08 | $6.48 |
| Kimi K2.6 | Agentic open model | $1.03 | $4.32 |
| DeepSeek V4 Pro | Deep reasoning value | $1.88 | $3.76 |
| Claude Sonnet 5 | Frontier coding + agents | $3.24 | $16.20 |
| Claude Opus 4.8 | Premium reasoning | $5.40 | $27.00 |
| Gemini 3.5 Flash | High-quality multimodal | $1.62 | $9.72 |
| Gemini 3.1 Pro | Frontier multimodal | $2.16 | $12.96 |
| Gemini 3 Flash | Fast multimodal | $0.540 | $3.24 |
| Gemini 3.1 Flash-Lite | Cost-efficient steps | $0.270 | $1.62 |
GPT-5.6 tiers, Kimi K2.6, and DeepSeek V4 Pro are served from Edgaze's managed Azure AI Foundry capacity on hosted runs. Claude models are served directly from Anthropic, Gemini models directly from Google.
Image Models#
| Model | Best for | Per image (standard) |
|---|---|---|
| Nano Banana 2Recommended | Default image generation | $0.072 |
| Gemini 3 Pro Image | Premium image quality | $0.145 |
| Nano Banana 2 Lite | High-volume, low cost | $0.037 |
| OpenAI Image 1 mini | Cheap, quick generations | $0.022 |
| ChatGPT Image 2 | Premium OpenAI image | $0.086 |
| OpenAI Image 1.5 | Balanced quality + cost | $0.065 |
OpenAI image rates vary with size and quality settings; the estimator prices the configured variant.
Embedding Models#
| Model | Best for | Per 1M input tokens |
|---|---|---|
| text-embedding-3-smallRecommended | Semantic search + RAG | $0.022 |
| text-embedding-3-large | Premium RAG retrieval | $0.140 |
Bring Your Own Key (BYOK)#
BYOK runs call the provider directly on your API key. OpenAI-brand, Anthropic, and Google models support this, and the token cost lands on your provider account instead of the rates above (a flat orchestration fee applies; see BYOK). Kimi K2.6 and DeepSeek V4 Pro run exclusively on Edgaze's managed capacity and do not support BYOK.
Automatic Model Migration#
Providers retire models on their own schedules, and the Edgaze catalog evolves with them. You never need to hand-edit old workflows:
- Selectable models appear in the builder dropdown and are fully supported.
- Deprecated models disappear from the dropdown but keep running and billing at their listed rate. Existing workflows are unaffected.
- Retired models are automatically routed to their designated successor within the same provider family at run time (for example, `gpt-4o-mini` → GPT-5.6 Luna, `claude-sonnet-4-6` → Claude Sonnet 5, `dall-e-3` → OpenAI Image 1.5). Cost estimates, holds, and billing all use the successor's rate, so what you're quoted always matches what actually runs.
Stored workflows are never bulk-edited: the remap happens when a run executes, and the builder writes the current id the next time you save a workflow.
Reliability#
Every chat and image model carries an ordered cross-provider failover chain. If a provider returns a transient error (rate limit, timeout, outage) mid-run, the engine substitutes the next model in the chain, capped by your run's cost hold, so failover can never make a run more expensive than what was quoted.
Questions#
Contact [email protected] for pricing questions, or see Pricing & Limits for account-level limits.
Related policies
Back to Legal & Trust CenterHow Edgaze estimates hosted compute at the input envelope maximum to set buyer displayed price, without deducting compute from creator earnings.
Per-run hosted workflow execution, BYOK behavior, funding sources, and abnormal usage controls.
Minimum and maximum price rules for prompt products and per-run workflow margin on Edgaze.