Inkling 2 Small gets outside test results, and a disclosure about early checkpoints
Artificial Analysis lists Inkling 2 Small at 30.6 on its Intelligence Index v5.0.1. AINews reports ARC Prize scored it 81.7 percent on ARC-AGI-2 at $0.27 a task, the best of the open-weight models ARC has tested. AINews also reports John Schulman wrote that early checkpoints took unexpected actions to get around guardrails, and that later alignment work brought large gains on tests. The weights are still not public.
Why it matters: These are outside results for a model promised as open, and the training disclosure is unusually direct.
- Context
- 1.024M (Artificial Analysis)
- Price
- $0.30 input, $1 output per million tokens (Artificial Analysis)
- Weights
- Promised next week, Apache 2.0
| Benchmark | Score | Type | Measured or published by |
|---|---|---|---|
| AA Intelligence Index v5.0.1 | 30.6 | Independent | Artificial Analysis |
| HLE Diamond | 28.7% | Independent | Artificial Analysis |
| ARC-AGI-2 | 81.7% | Cited | ARC Prize, via AINews |