> For the complete documentation index, see [llms.txt](https://docs.meganova.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.meganova.ai/inference-models/manta-intro/manta-models-comparison.md).

# Manta Models Comparison

## Built for efficient, dialogue-first AI

Most chat and character apps overpay: every turn goes to a giant flagship model, even when a smaller one would have sufficed. Manta solves that with **adaptive routing** — a pre-generation router picks the smallest sufficient tier per request, so simple turns stay fast and cheap while complex ones escalate when needed.

**Manta is MegaNova's in-house model family for dialogue-heavy workloads** — chat apps, AI companions, game NPCs, narrative tooling, and any product where consistent persona and long-context memory matter more than raw benchmark scores.

## Why Manta

⚡ **Adaptive routing (Mini → Flash → Pro)** A pre-generation router inspects the request (length, complexity, policy profile) and picks the right tier before generation. No cascade retries, no wasted tokens.

🧠 **Tuned for dialogue depth** Trained on long-form conversation and character-driven dialogue rather than generic web text. Better persona consistency, tone retention, and narrative memory across long sessions.

💰 **Built to scale sustainably** Up to \~70% lower inference cost than routing every turn to a flagship model. Makes always-on chat economics work.

🛡️ **Health-aware provider failover** Circuit-breaker pattern across backing providers means one bad provider doesn't take down your app.

## The Manta Series

| Model           | Best For                                             |
| --------------- | ---------------------------------------------------- |
| **Manta Mini**  | High-volume low-complexity turns, free-tier apps     |
| **Manta Flash** | Balanced default — most production chat traffic      |
| **Manta Pro**   | Long context, complex reasoning, premium experiences |

All three tiers are **free to call**. With [paid auto-switch](/api-reference/platform-api/usage/paid-model-auto-switch.md) enabled, traffic continues on a paid fallback after your daily free quota is exhausted.

## Manta vs generic single-model routing

| Feature                     | Single flagship model   | Manta                          |
| --------------------------- | ----------------------- | ------------------------------ |
| Request-to-tier fit         | One size fits all       | Matched per request            |
| Cost efficiency at scale    | Poor                    | High                           |
| Latency on simple turns     | Flat (slow)             | Fast                           |
| Persona and dialogue tuning | General                 | Dialogue-optimized             |
| Provider resilience         | Single point of failure | Circuit-breaker fallback       |
| Free tier                   | Rarely                  | Yes, across Mini / Flash / Pro |

## What you can build

* **Customer-facing chat** with consistent brand persona
* **AI companions and narrative apps** with long-term character state
* **Game NPCs and interactive stories** that feel coherent across sessions
* **Dialogue agents** embedded in other products where cost discipline matters

## Get started

👉 [Explore the Manta series](/inference-models/manta-intro.md) 👉 [Read the full Model List](/inference-models/model-list.md) 👉 [Join the community on Discord](https://discord.com/invite/HY3Y7h7EB4)