asi1-ultra: frontier intelligence, the lowest cost of outcome
Ultra builds on GLM-5.3, an open-weights model built for complex reasoning, coding, and long agentic runs. We benchmark, compare, and validate it across our own agent harnesses before it ships as asi1-ultra - then we work with your team to tune and train the model to meet your requirements, so you never spend time researching models or swapping them out.
Intelligence
60
Tied for the top open-weights model on the Artificial Analysis Intelligence Index, 3 points behind Claude Opus 5 and 8th of 181 models evaluated.
Cost of outcome
$0.68
The lowest cost per Intelligence Index task of any model scoring 58 or above. Claude Opus 5 reaches its score at $2.34 per task - 3.4x the cost.
How it compares
Artificial Analysis is the independent evaluator of AI model intelligence, performance, and pricing. Their Intelligence Index aggregates ten evaluations across reasoning, knowledge, mathematics, and coding; cost per task is the weighted average USD cost to run one Intelligence Index task.
How to read it: each dot is one model. Dots higher up score better on the Artificial Analysis Intelligence Index; dots further left cost less per task. The best cost of outcome sits up and to the left.
| Model | Intelligence Index | Cost per task |
|---|---|---|
| GLM-5.3 (asi1-ultra) | 60 | $0.68 |
| Kimi K3 | 60 | $0.84 |
| Claude Opus 5 | 63 | $2.34 |
Scores from Artificial Analysis, the independent evaluator. Intelligence Index scores change with new evaluation versions - verify current scores before citing.
What it does best
- Complex multi-step reasoning and long agentic runs
- Coding: tops open-source benchmarks for terminal work and coding agents
- Document analysis with precise citations
- The deepest thinking mode in the asi1 family
We do the benchmarking. You get the outcome.
Part of how ASI:One works: we benchmark every model against the tasks your team actually runs - long tool chains, document analysis, code review - before it ships behind a tier. Then we work with you to tune and train the model to your requirements, so you never spend time researching models, comparing options, or upgrading and swapping them out as the market moves.
Try it in ASI:WorkVerify the latest scores on Artificial Analysis - Artificial Analysis is the independent evaluator of AI model intelligence, performance, and pricing. Their Intelligence Index aggregates ten evaluations across reasoning, knowledge, mathematics, and coding; cost per task is the weighted average USD cost to run one Intelligence Index task..