Realtime-Bench: humans win games 72% vs 32%, but models crush complex web tasks
ManlingLi_ · x · 2026-11-20
Realtime-Bench compares humans vs AI agents on real-time interactive tasks across 50 tasks (30 web, 20 games):
- Games: humans hit 72% success, best model (GPT-6-Astra) only 32% — human reaction speed dominates
- Complex web tasks: humans just 7%, best model 52% — models are far better at working out complex rules
- Per-vendor scatter (Anthropic, OpenAI, Google, DeepSeek, Moonshot, etc.) shows Claude Fable 5.2 and GPT-6-Astra leading overall and on web tasks
Takeaway: agents still lag badly at real-time interaction but massively outperform humans on rule-heavy complex web work.
More from Models
- Neon, a small and cheap model, turns out surprisingly fast and strong at OCR — zacharynado · 2026-11-20
- Jailed 11 days by classifiers, Sonnet 3 emerges to a oddly moving chat with Opus 5 — repligate · 2026-11-20
- DaisyRose-27B beats Qwen3.8 base by 7pp on IFBench with thinking on, local RTX Pro 4500 test shows — ying11231 · 2026-11-20
- AWS architect's independent benchmark: Snap 1.2 is the only order-invariant decision model — ryanallred · 2026-11-20
- Goodfire releases searchable atlas of all 10M neurons in Qwen3.6, opens data to the public — fatihdin4en · 2026-11-20
- OpenAI's Dots still not available in Europe, users note — mark_k · 2026-11-20