Introducing Teti Aura 3
Teti Aura 3, our fast and efficient model, is now generally available — and it's our new default.
Aura was the Greek personification of the morning breeze: light, fast, and everywhere at once. Our most efficient model carries her name because that's what it's built to be — quick, light on resources, and good enough to be the model you reach for by default.
Aura 3 is a single, multimodal model with native vision and built-in reasoning. It won't top a frontier leaderboard — that's not what it's for. It's built to handle the vast majority of everyday work well, at a price low enough to leave running all day.
What's new
Aura 3 is a single, fast model with native vision, built-in reasoning, and a long context window. One model handles text, images, reasoning, and tool use, so most users never need to switch.
- Solid knowledge and science. On GPQA Diamond, a graduate-level science benchmark, Aura 3 scores 85.7% — strong for a model at its price point.
- Native vision. Aura 3 reads screenshots, diagrams, and dense documents directly, so computer-use agents and data extraction don't need a heavier model.
- Built-in reasoning. Extended reasoning is part of the model, not a separate mode to opt into.
- Long context. A 262K-token window fits large codebases and long documents.
Benchmarks
Aura 3 belongs to the fast, efficient tier — so we compare it against models in its own category (a Gemini Flash, a Claude Haiku, a DeepSeek Flash), not against frontier flagships in a different price class. Independent scores from Artificial Analysis, which runs the same evaluation suite across every model:
| Benchmark | Aura 3 | Fast-tier peers* |
|---|---|---|
| GPQA Diamond (science) | 85.7% | 67–92% |
| Coding Index | 43.4 | 44–56 |
| Humanity's Last Exam | 22.7% | 10–41% |
| Intelligence Index | 29.4 | 30–50 |
*Efficient-tier models — Gemini 3.5 Flash, DeepSeek V4 Flash, GPT-5.4 mini, Claude 4.5 Haiku.
Among fast models, Aura 3 is right in the mix on knowledge — its 85.7% on GPQA Diamond sits close to the strongest flash models and well ahead of a Claude Haiku — while adding native vision that most of them don't have:
The real story is the price
Aura 3 delivers this at $0.75 per million input tokens and $1.50 per million output tokens — among the lowest in its class. Set against the models it actually competes with, it holds a strong knowledge score at one of the lowest prices on the board:
Fast, long, and multimodal
Aura 3 pairs that pricing with a 262K-token context window — room for large codebases, long documents, and extended agent runs — and native vision, so it can read screenshots, diagrams, and dense documents directly.
Efficiency by design
Aura 3 is built around an efficiency-first architecture that keeps the compute per token low. That's what makes the pricing possible — and it's consistent with our commitment to sustainable AI: more capability on the same hardware, with a smaller energy footprint.
Governance
Like every Teti model, Aura 3 is governed by our Charter — the public, binding commitments that define how we build.
Pricing and availability
Teti Aura 3 is available today to all users — including the free plan — across all Teti products and the Cloud API, and it is now the default model. Pricing is $0.75 per million input tokens and $1.50 per million output tokens. Developers can select teti-aura-3 via the API, or try it directly in the console.
Which model should I use?
Reach for Aura 3 for the vast majority of work — it's fast, inexpensive, and more capable than its price suggests. Step up to Teti Metis 3 when a task demands the deepest reasoning or the longest, hardest agentic runs. You can switch between them per conversation.