Persimmon Pro/Ultra launches with test-time scaling, doubling Multi-User Turing Test score to 48%
niloofar_mire · x · 2026-11-21
humansand released Persimmon v0.1 Pro and Ultra, two new tiers using test-time scaling to make each conversation turn more realistic, now open to everyone without gated access.
Ultra more than doubles Persimmon's score on the 8-turn Multi-User Turing Test, from 18% to 48% — at 50%, a model becomes indistinguishable from real humans under their held-out judge.
The team notes that test-time scaling helps user simulation capabilities too, crediting Charlie G for the Herculean engineering effort.
More from Models
- Claude's 'Riemann Hypothesis proof' unravels: a 250,000-page slop book and a tautological Theorem 1 — basedjensen · 2026-11-21
- GPTZero scans millions of citations: hallucination rates flat or rising at CVPR, ICLR, EMNLP — HanchungLee · 2026-11-21
- Investor slams per-token pricing: verification burns millions of tokens, calls for pay-per-outcome — jasondeanlee · 2026-11-21
- Custom RTS benchmark for agents: GPT-6.2 Sol scores 98.9, crushes field — asimpleabstraction · 2026-11-21
- 4,700-vote marketing eval: GPT-6.2 Sol beats Opus 5.5 at 58% win rate and a quarter of the cost — boringmarketer · 2026-11-21
- DIY RTS benchmark for agents: GPT-6.2 Sol posts 98.9 win rate, tops all rivals — asimpleabstraction · 2026-11-21