Models & pricing

The catalog, the providers behind it, fallback chains, and how per-token pricing works.

The catalog

Conduix curates a frontier catalog of production-grade models across leading providers — every id pinned and price-verified against the provider (never floating "-latest" aliases). Browse the full catalog — searchable, filterable, with per-1M-token pricing — at /models (no login required), or see your org's live view at /dashboard/models.

OpenAI: gpt-5.5, gpt-5.4, gpt-5.4-mini, gpt-5.4-nano, gpt-5.3-codex
Anthropic: claude-fable-5, claude-opus-4-8, claude-sonnet-5, claude-haiku-4-5
Google: gemini-3.1-pro-preview, gemini-3.5-flash, gemini-3.1-flash-lite
xAI: grok-4.3
DeepSeek: deepseek-v4-flash, deepseek-v4-pro
Mistral (EU): mistral-medium-2604, mistral-large-2512, mistral-small-2603, codestral-2508
Groq: gpt-oss-120b, gpt-oss-20b, Llama 4 Scout (fastest inference)
Together: Qwen 3.7 Plus, Kimi K2.6, Kimi K2.7 Code, GLM 5.2, Gemma 4
Fireworks: DeepSeek V4 Flash, Qwen 3.7 Plus, Kimi K2.7 Code, GLM 5.2 (redundancy routes)
Azure AI Foundry: DeepSeek V4 Pro (US-hosted on Microsoft-managed infra)
Amazon Bedrock: Amazon Nova Pro

Routing aliases — skip model names entirely

Don't want to track model releases? Request a capability instead. Aliases are valid model values that resolve to a curated cross-provider chain, honoring your org's routing policy (residency, provider/model blocks, PHI allowlists) per candidate. The response tells you what served via x-conduix-alias and x-conduix-model-served.

General: frontier-best · frontier-fast · frontier-budget · enterprise
Coding: code-premium · code-balanced · code-fast · code-open · agentic-code
Reasoning: reasoning-best · reasoning-fast · reasoning-budget
Vision & documents: vision-best · vision-fast · document-ai
Coming soon: embeddings · image · speech (stt/tts) · video · ocr · rerank — ship with their endpoints

python

client.chat.completions.create(
    model="frontier-best",  # → resolves to the best available frontier model
    messages=[{"role": "user", "content": "…"}],
)

How pricing works

Transparent per-token pricing. The rate at /dashboard/models is what you pay — all-in. Credits behave like dollars (1 credit ≈ $1). No per-seat fees, no surprise overages.

Pay-as-you-go: Buy credits in advance, spend as you go
Auto-reload: Top up automatically when you hit a threshold
Subscription plans: Explorer, Builder, Scale — see /pricing
Volume discounts: Custom unit pricing for committed volume
Enterprise: NET-30 invoice, POs, custom MSAs

Calling a model

Use the model id in the standard OpenAI request body. Conduix routes to the correct provider based on the id.

python

client.chat.completions.create(
    model="claude-haiku-4-5-20251001",  # → routed to Anthropic
    messages=[{"role": "user", "content": "Summarize…"}],
)

Automatic fallback chains

Each model has a curated fallback list. If the primary fails (provider down, circuit breaker tripped, rate-limited upstream), Conduix retries against the next entry — invisibly. Your customer never sees the failure.

text

Requested:  gpt-5.4-mini
Fallback chain:
  1. gpt-5.4-mini         (openai)
  2. claude-haiku-4-5     (anthropic)
  3. gemini-3.1-flash-lite (google)

Retired models never 404

When a provider retires a model, Conduix removes it from the catalog and transparently remaps requests to its successor — your integration keeps working. The response carries x-conduix-model-remapped: old-id -> new-id and the substitution is recorded in your audit log.

When a fallback served the request, the response includes x-conduix-fallback: true and x-conduix-model-served shows the actual model used. See Observability.

Restricting which models a key can call

Per-key allowlists keep developers in their lane (e.g. internal tools can't accidentally call frontier-tier models). Configure in the dashboard or via the keys API:

json

{
  "name": "Internal tools",
  "allowed_models": ["gpt-5.4-mini", "frontier-budget"],
  "rate_limit": 60
}

Allowlists accept concrete model ids and routing aliases — granting frontier-budget entitles the key to whatever that alias resolves to under your org policy, without naming models.

Using a non-catalog model

Need a model Conduix doesn't curate? Or your own self-hosted model? Use BYO endpoints: register your own OpenAI-compatible URL and call model: "byo:<slug>/<your-model>".

Tiers

Models are grouped into tiers for governance. You can build allowlists at the tier level rather than naming each model:

frontier: Newest, top-performing, most expensive (e.g. gpt-5.5, claude-fable-5, claude-opus-4-8)
premium: Strong general-purpose (gpt-5.4, claude-sonnet-5, deepseek-v4-pro, Kimi K2.6)
mid: Cost-effective workhorses (gpt-5.4-mini, claude-haiku-4-5, gemini-3.5-flash)
budget: Best value — open-weight workhorses (Groq gpt-oss, DeepSeek V4 Flash) for high-volume, routine work

Authentication Billing & credits