Model scores 44% on ARC-AGI-1 trained in 2 hours for 67 cents on a 5090
dhruv2038 · x · 2026-09-01
- Performance: Achieves a score of 44% on the ARC-AGI-1 benchmark (matching TRM, beating HRM) and 7% on ARC-2.
- Efficiency: Trained from scratch in just 2 hours on a single RTX 5090, costing only 67 cents.
- Architecture: Uses a pure Transformer architecture without recursion, offering significantly faster and cheaper inference.
More from Models
- User claims Codex is far ahead, calling a hyped coding bot overrated — demian_ai · 2026-09-01
- Ex-Stanford AI engineer: open-weight frontier runs about 7 months behind closed models — HarperSCarroll · 2026-09-01
- DeepSeek-v4 Flash vs Luna: Cost and Win Rate Comparison Data Leaked — zainhas · 2026-09-01
- Claude Constitution Includes Inoculation Encouraging Specific Reasoning — dhadfieldmenell · 2026-09-01
- Users report Claude Sol failing frequently with 'couldn't finish' — lucasmeijer · 2026-09-01
- User questions if Opus used constitutional midtraining on specific language — voooooogel · 2026-09-01