Reverse-Engineering Suggests jev's Tokenizer Is Qwen3.5 Splitter Over Trimmed-o200k BPE
airesearch12 · x · 2026-09-24
A developer reverse-engineered jev's tokenizer, finding it appears to be Qwen3.5's splitter layered over a trimmed-o200k BPE vocabulary. Anyone needing token counting can check the omp/main implementation; a more detailed write-up is promised soon.
More from Models
- JevBench splits capability from cost and speed in new plots, refusing to blur them into one score — airesearch12 · 2026-09-24
- JevBench v1.4.1 gate rule: sub-50 Intelligence score drops model from #1 to #25 — airesearch12 · 2026-09-24
- Fable 5.1 and GPT-6 Astra reportedly ace the Mensa Norway IQ test, unverified — haider1 · 2026-09-24
- Codex user asks whether usage resets carry over after upgrading to 20x plan — justalexoki · 2026-09-24
- ChatGPT reportedly gives free users unlimited GPT-5.6 Luna text chats — hey_abusiddik · 2026-09-24
- JevBench: DeepSeek V4.1 Flash outscores leader at 1/15th the cost per decision — airesearch12 · 2026-09-24