Capability.
A live tracker of frontier AI progress across the six dimensions that compose the path to ASI. Reasoning, autonomy, multimodal, planning, alignment, composability. Calibrated quarterly against a reference benchmark suite and a 30-expert review panel. Open methodology. Adopt it.
All frontier models.
95+ on all six dimensions through composition. The frontier sits 42 points away on aggregate — with the largest gaps in autonomy horizon and long-range planning. Composed networks have already crossed 78 on composability and 67 on alignment, suggesting the path forward is architectural, not scale.
Every frontier model, every dimension.
Published openly under CC BY 4.0. Any lab may adopt this protocol for internal benchmarking. Replication corrections welcome at support@fasciaai.com.
| Model | Reasoning | Autonomy | Multimodal | Planning | Alignment | Composability | Aggregate |
|---|---|---|---|---|---|---|---|
| GPT-5 (Sep 2025) | 67.1 | 40.2 | 74.5 | 44.3 | 57.2 | 49.8 | 55.5 |
| Claude 4 Opus | 65.8 | 42.1 | 69.3 | 45.8 | 61.0 | 50.2 | 55.7 |
| Gemini 3 Ultra | 63.4 | 36.7 | 73.9 | 40.2 | 53.4 | 46.5 | 52.4 |
| Llama-4 405B (Open) | 58.9 | 32.0 | 65.8 | 36.1 | 48.5 | 43.2 | 47.4 |
| Frontier mean | 64.2 | 38.4 | 71.1 | 42.0 | 54.8 | 47.9 | 53.0 |
| Composed (5-agent) | 69.2 | 56.3 | 72.4 | 58.1 | 67.5 | 78.4 | 67.0 |
Get the full dataset.
Quarterly benchmark data, per-model breakdowns across all 240+ benchmark tasks, reviewer panel annotations, 60-day early access to forthcoming volumes, and a private channel to the research team. For frontier labs, AI VCs, F500 CIOs.