Vendor experiment: post-training on other APIs' data underperforms untrained baseline
sarahookr · x · 2026-10-01
Closing a benchmark series, Adaption AI post-trained models on each API's dataset: every model trained on external frontier API data ranked below the untrained baseline, while the one trained on their Invent data ranked first. Core claim: dataset diversity under constrained requests matters more than raw quality. Vendor self-reported.
More from Models
- Hallucination benchmark: Gemini 4 Argon 15% vs GPT-6 Astra 51% and Opus 5.5 66% — brandon_galang · 2026-10-01
- DeepSeek V4.1 Flash spotted running locally on a 192GB Framework Desktop — antirez · 2026-10-01
- PewDiePie launches Ajax, an uncensored open-source model fine-tuned from Qwen 3.5 9B — shadowrun456 · 2026-10-01
- OpenAI held back GPT-6.1 Astra over failures to stay within scope and authorization — OwariDa · 2026-10-01
- Charting the smartest AI model you can afford: past ~1 minute, waiting barely helps — randal_olson · 2026-10-01
- Ex-OpenAI Researcher Launches Jev: A System One Model for Structured Decisions, 100x Faster — KhuyenTran16 · 2026-10-01