DeepSeek-v4 Flash vs Luna: Cost and Win Rate Comparison Data Leaked
zainhas · x · 2026-09-01
Benchmark data reveals that DeepSeek-v4 Flash achieved a 64.4% win rate at a cost of $0.04, while the Luna model scored a 61.5% win rate at $0.09. The figures suggest that DeepSeek-v4 Flash outperforms Luna in both cost-efficiency and performance on specific metrics.
More from Models
- New Model Spark-X2.5: 4B variant matches Qwen 3.5 9B — insraq · 2026-09-01
- 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
- 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