Opus finally getting BETTER?! (SpaceX Deal)

Anthropic lands a SpaceX compute deal — 300 megawatts — but Opus 4.7 is still not as smart as the old 4.6.

What changed for users

Effective immediately, Anthropic is "substantially" raising API rate limits and doubling the 5-hour rate limit for Pro users. They are also removing peak-hour limit reductions. The weekly cap is unchanged, so burning through the longer 5-hour window still eats into the weekly allowance. No new tokenizer increase, and Opus 4.5 is gone — leaving 4.6 and 4.7 as the only options.

Does it actually help?

The creator ran tests the same day and concluded: "I don't see Opus getting smarter." The SpaceX deal, 300 MW of extra compute, is real capacity, but it does not reverse what users have been calling "nerfs" — repeated degradations that pushed the creator to label Opus as "among the commoners" rather than the "orange-headed giant" it used to be. GPT, Gemini, and Kimi are now described as "awfully good."

What the creator switched to

Two to three weeks ago, the channel moved agent work, coding, monitoring, and scraping off Claude and onto Kimi K2 + Hermes agent running in a TUI, which shows grep calls from the agent swarm. A shared Kimi code across roughly five users has only burned 18% of quota, versus repeated Claude Max plan exhaustion under the old Opus workload.

When to still use Opus

Use Claude for "hyper difficult tasks" as long as you explicitly tell it to fix its own mistakes — laziness and blame-shifting (e.g. claiming TypeScript "became more strict in the last 20 minutes") remain common. The creator's rule: don't touch Claude again "until the nerf is not nerfed."

How to read the deal

The 300 MW SpaceX partnership confirms a compute shortage, not a capability upgrade. The bottleneck is hardware, and everyone — Anthropic included — is competing for the same Nvidia GPUs.