Grok-4.8

Grok 4.8 reportedly taking longer than planned in RL as OpenRouter spending on OpenAI and Anthropic rises

According to a post on X by @ramez, OpenRouter data shows spending on both OpenAI and Anthropic models rising sharply over 12 weeks, tied to the Opus 5.5 and GPT-6 Astra launches. @ramez reads this as users paying for, and quickly switching to, the best model. The same aggregated feed cites @koltregaskes saying Grok 4.8, described as a 2.5 trillion parameter model, finished training in mid-September but is taking longer than planned in RL. The provided snippet is cut off before the remaining posts, including one from @thdxr about his team.

Why it matters: For developers and businesses choosing models, the reported spending shift suggests usage can move quickly toward whichever model is newest and strongest. Teams relying on Grok 4.8 may want to watch whether the RL timeline changes. The headline also mentions builders warming to Gemini, but the excerpt available here does not include details on that point.