❱ Models

The Majestix AI Inference Hub provides access to 18 models across 4 providers: Anthropic, OpenAI, Vertex AI (Google), and OpenRouter. All models are accessed through a single unified API endpoint (POST /chat) using a consistent request format and SSE streaming protocol.

Available Models

Key	Provider	Underlying Model	Context Window	Best For
claude-sonnet	Anthropic	Claude Sonnet 4.6	200K	Balanced performance across all tasks
claude-haiku	Anthropic	Claude Haiku 4.5	200K	Fast responses, cost-sensitive workloads
claude-opus	Anthropic	Claude Opus 4.6	200K	Complex reasoning, difficult problems
gpt-5-mini	OpenAI	GPT-5 Mini	400K	Fast general-purpose tasks
gpt-5.2	OpenAI	GPT-5.2	400K	Flagship-quality responses
gpt-5.2-codex	OpenAI	GPT-5.2 Codex	400K	Code generation, agentic workflows
cmo-agent	OpenAI (fine-tuned)	CMO Agent	128K	Marketing content and strategy
gemini-3-flash	Vertex AI	Gemini 3 Flash	1M	Fast responses with massive context
gemini-3.1-pro	Vertex AI	Gemini 3.1 Pro	1M	Reasoning with massive context
gemini-3-image	Vertex AI	Gemini 3 Image	--	Image generation
gpt-5-image	OpenAI	GPT-5 Image	--	Image generation
seedream-4.5	OpenRouter (Google)	Seedream 4.5	--	High-quality image generation
llama-4-maverick	OpenRouter (Meta)	Llama 4 Maverick	1M	Large-context open-source tasks
deepseek-v3.2	OpenRouter	DeepSeek V3.2	164K	Budget-friendly general tasks
deepseek-r1	OpenRouter	DeepSeek R1	64K	Reasoning and chain-of-thought
qwen3-coder	OpenRouter	Qwen3 Coder	262K	Code generation, agentic workflows
kimi-k2.5	OpenRouter (Moonshot)	Kimi K2.5	262K	Versatile general-purpose tasks
grok-4.1-fast	OpenRouter (xAI)	Grok 4.1 Fast	2M	Tasks requiring the largest context window

Custom Models

The platform includes proprietary fine-tuned models exclusive to Majestix AI, identified by the is_custom flag in the model catalog:

Model	Specialization	Details
cmo-agent	Marketing & Strategy	Fine-tuned OpenAI model trained as a Chief Marketing Officer. Excels at campaigns, copywriting, SEO, brand strategy, and go-to-market planning. See OpenAI Models for full details.

Custom models are tagged with a Custom badge in the web app's model selector. More custom models are planned for future releases.

---

Free Tier Model Access

The Free plan ($0/month, 500 credits) restricts model access to fast-tier models only:

Available on Free	Not Available on Free
claude-haiku	claude-sonnet, claude-opus
gpt-5-mini	gpt-5.2, gpt-5.2-codex, cmo-agent
gemini-3-flash	gemini-3.1-pro
deepseek-v3.2	deepseek-r1, qwen3-coder, kimi-k2.5, grok-4.1-fast, llama-4-maverick
—	All image generation models

Upgrade to Guru ($10/month) or Pro ($50/month) for full access to all 18 models. See Credits & Billing for plan details.

---

Provider Pages

• Anthropic Models -- Claude Sonnet, Haiku, and Opus

• OpenAI Models -- GPT-5 family and CMO Agent (custom)

• Vertex AI Models -- Gemini family and Seedream

• OpenRouter Models -- Llama, DeepSeek, Qwen, Kimi, and Grok

---

Auto-Routing

If you omit the model parameter from your POST /chat request (or select Auto Router in the web app), the platform automatically selects the best model for your task. The router considers:

• Message content — marketing queries route to cmo-agent, coding queries to coding-optimized models

• Task complexity — simple Q&A routes to fast models, complex analysis routes to reasoning models

• Message length — very long inputs route to large-context models (Gemini, Grok)

• Model strengths — each model is tagged with specializations that inform routing decisions

Auto-routing respects your plan's model access. Free-tier users are routed within the fast-tier models only.

Example -- explicit model selection:

{
  "model": "claude-sonnet",
  "messages": [{"role": "user", "content": "Explain quicksort."}]
}

Example -- auto-routed (no model specified):

{
  "messages": [{"role": "user", "content": "Explain quicksort."}]
}

Provider Fallback

When a primary provider experiences an outage or returns a transient error, the platform automatically retries the request on a comparable model from a different provider. This happens transparently -- the client receives a normal streamed response and does not need to handle retry logic.

The fallback map pairs models with similar capability profiles across providers. For example, a request targeting claude-sonnet may fall back to gpt-5.2 or gemini-3.1-pro if Anthropic is unreachable. Fallback is attempted once; if the fallback provider also fails, the error is returned to the client.

Fallback routing is designed to preserve response quality. A flagship model will never fall back to a budget model or vice versa.

Token Limits and max_tokens Clamping

Each model defines a maximum output token limit. If your request specifies a max_tokens value that exceeds the model's supported maximum, the platform automatically clamps it down to the model's limit rather than rejecting the request. This ensures that requests are never refused solely due to an overly large max_tokens value.

If you do not specify max_tokens, the platform uses a sensible default for the selected model.

Choosing a Model

Use the following guidelines to select the right model for your use case:

Use Case	Recommended Models
General chat and Q&A	claude-sonnet, gpt-5.2, kimi-k2.5
Fast, cost-effective responses	claude-haiku, gpt-5-mini, deepseek-v3.2, gemini-3-flash
Complex reasoning and analysis	claude-opus, deepseek-r1, gemini-3.1-pro
Code generation and agentic tasks	gpt-5.2-codex, qwen3-coder
Very large documents (100K+ tokens)	gemini-3-flash, gemini-3.1-pro, llama-4-maverick, grok-4.1-fast
Marketing content	cmo-agent
Image generation	gemini-3-image, gpt-5-image, seedream-4.5