Introducing Teti Metis 3
Our most capable model, Teti Metis 3, is now generally available.
In Greek mythology, Metis was the Titaness of deep thought and wisdom — the counsel the gods turned to when a problem had no obvious answer. Our most powerful model carries her name because that is exactly what it's built for: the hard problems that reward patience, rigor, and reasoning that holds together over a long stretch of work.
Metis 3 is a major step up from Aura on the tasks that matter most for professional work: complex reasoning, multi-step research, and agentic execution that has to stay coherent across many steps. It pays precise attention to instructions, plans before it acts, and verifies its own outputs before reporting back.
What's new
Metis 3 is a native multimodal model with reasoning and vision built in. One model handles text, images, deep reasoning, and tool use — no switching, no separate "thinking" variant to opt into.
- Near-frontier knowledge. On GPQA Diamond, a graduate-level science benchmark, Metis 3 scores 91.1% — within a few points of the best closed frontier models (92–94%).
- Stronger reasoning and coding than Aura. A higher Intelligence Index (42.8) and Coding Index (56.0) make it the model to reach for on harder, longer work.
- Native vision. Reads diagrams, screenshots, and documents in detail, no separate vision model required.
- Long context. A 256K-token window for large codebases, long documents, and extended agent runs.
Benchmarks
Metis 3 competes with the frontier flagships, so that's the category we compare it against — not the fast, cheaper tier where Aura sits. Independent scores from Artificial Analysis, which runs the same evaluation suite across every model:
| Benchmark | Metis 3 | Frontier reference* |
|---|---|---|
| GPQA Diamond (science) | 91.1% | 92–94% |
| Coding Index | 56.0 | 69–75 |
| Humanity's Last Exam | 35.9% | ~45% |
| Intelligence Index | 42.8 | 47–56 |
*Best current closed frontier models — Gemini 3.1 Pro, GPT-5.5, Claude Opus 4.8.
On knowledge-heavy reasoning, Metis 3 is genuinely in the frontier conversation — its 91.1% on GPQA Diamond sits just behind models that cost five to six times as much. On the broadest composite tests it trails the frontier, and we say so; the trade it makes is near-frontier capability at a fraction of the price.
Frontier-tier work without the frontier price
Metis 3 reaches that tier at $2.00 per million input tokens and $5.00 per million output tokens — a fraction of the leading closed models. It's the model to reach for on hard problems without the frontier price tag.
Built for long, hard work
Where Metis 3 separates itself is on tasks that don't fit in a single answer. It keeps track of context across long sessions, knows when to gather more information instead of guessing, and recovers gracefully when an approach stops working. Metis 3 supports a 256K-token context window and up to 33K tokens of output, with native vision for reading diagrams, screenshots, and documents in detail.
Governance
Metis 3 ships behind the same commitments as the rest of Teti — governed by our Charter and Ethics framework, the public rules that define how we build.
Pricing and availability
Teti Metis 3 is available today to Pro, Max, and Quantum users across all Teti products and the Cloud API. Pricing is $2.00 per million input tokens and $5.00 per million output tokens, with cached input at $0.35 per million. Developers can select teti-metis-3 via the API, or try it directly in the console.
Migrating to Metis 3
Metis 3 is a direct upgrade for anyone doing complex reasoning, research, or engineering work. Because it reasons more before answering, it can produce more output tokens on hard problems — the tradeoff is higher reliability where it counts. If you need lower latency and cost for everyday tasks, Teti Aura 3 remains the better fit, and you can move between the two per task.