ZMIME
AuroraAI Research · Released Nov 18, 2026

Aurora2-400M-A20M

English base model pretrained on 100 billion tokens with one shared expert and 32 routed experts per layer.

  • Open weights
Capability
—No ECI score yet
Input
—No listed price
Output
—No listed price
Context
—tokens
Pricing

What it costs

No per-token price is published for this model yet.

No per-token price is listed for this model on models.dev.

Capability

Independent benchmarks

Epoch AI has not published a Capabilities Index score for Aurora2-400M-A20M yet.

Specifications

The details

Lab
AuroraAI Research
Task
Base language modeling
Published capabilities
English base model pretrained on 100 billion tokens with one shared expert and 32 routed experts per layer.
Model size
400M total / approximately 19M active parameters
Access
Public Transformers checkpoint for completion and adaptation
Released
Nov 18, 2026
Context window
— tokens
Inputs
Text
Output
Text
Reasoning
Unconfirmed
Tool calling
Unconfirmed
Structured output
Unconfirmed

AuroraAI Research model documentation

Questions

About Aurora2-400M-A20M

How much does Aurora2-400M-A20M cost?

Aurora2-400M-A20M has no published per-token API price on models.dev yet.

How good is Aurora2-400M-A20M?

Epoch AI has not published a Capabilities Index score for Aurora2-400M-A20M yet.

Is Aurora2-400M-A20M open source?

Yes. AuroraAI Research publishes the weights under the Apache-2.0 licence, so it can be downloaded and self-hosted.

What inputs does Aurora2-400M-A20M support?

Aurora2-400M-A20M accepts text. English base model pretrained on 100 billion tokens with one shared expert and 32 routed experts per layer.

When was Aurora2-400M-A20M released?

AuroraAI Research released Aurora2-400M-A20M on Nov 18, 2026.

What are the best alternatives to Aurora2-400M-A20M?

The closest current models from other labs on capability, price and release date are Hy4 flash Base (Tencent), Aion 1.0 Instruct 2.3B (Microsoft), DaisyRose-27B (RadixArk) and Olala Mini 7A1B (OVHcloud).