Why labs route APIs to Claude: SFT gives a boost, but capabilities come from RL and envs
xeophon · x · 2026-09-13
xeophon questions why some labs route their APIs or products to Claude, arguing it offers no technical advantage and hurts reputation. In follow-up, he adds technical context: while SFT gives a boost, capabilities these days come from RL and environments — hinting that some externally served models rely on Claude under the hood.
More from Models
- GPT-6 Astra builds interactive simulation from a single image, wowing researchers — akapoor_av8r · 2026-09-13
- Nex-N2.5 agentic models hit OpenRouter free tier, built around visual feedback loops — anthara_ai · 2026-09-13
- Dwarkesh Patel: Why Every AI Model Is Starting to Sound the Same — Dwarkesh Patel · 2026-09-13
- Paid €200 for Codex and can't use it: all five models refuse legitimate security triage — ClassicMain · 2026-09-13
- ChatGPT left a task hanging for 45 minutes, forcing the user to kill the thread — codepoetn · 2026-09-13
- Chris Paxton: still finding problems astra can't solve amid the hype — chris_j_paxton · 2026-09-13