
Crypto Révolution 🇫🇷
33.7K posts

Crypto Révolution 🇫🇷
@CryptoRevoltFR
🇫🇷 Décryptage, analyses & trading sur #Bitcoin et autres crypto I TOLD YOU SO maker since 2018






@levelsio @nfcodes I created a redis instance on hetzner with public port open for few minutes and someone was running a cryptominer the next moment taking 50% CPU 💀 After that I always use @Tailscale 👌




OpenClaw just got a lot cheaper to run. Alibaba Cloud dropped a Coding Plan that gives you 4 frontier models under one API key. Plug it straight into OpenClaw and you're done. Qwen 3.5-Plus. Kimi K2.5. MiniMax M2.5. GLM-5. 18,000 requests a month for just $10. In single subscriptions, you can swap the models and keep building. The interesting part isn't even the price. It's that they built this for developers already inside tools like OpenClaw, Claude Code, and Cline. The top engineers will quietly plug this in this week and say nothing. The rest will find out in 6 months when the cost gap is impossible to ignore. Setup takes 30 seconds. 👇

CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING does not "unlock" Opus 4.6 Here is what it actually does, and what you should set instead. There are THREE thinking-related settings in Claude Code. Here is what each one does: 1. effortLevel ("low" / "medium" / "high") Controls how deeply Claude reasons on each turn. At "high," Claude almost always uses extended thinking. At "medium," it thinks less aggressively. At "low," it may skip thinking entirely on simple tasks. On March 4th (v2.1.68), Anthropic changed the default effort from "high" to "medium" for Max and Team subscribers. This is a documented changelog entry. If your Claude Code suddenly felt less capable around that date, this is probably why. Not a nerf. Not a conspiracy. A default changed. Fix: set "effortLevel": "high" in ~/.claude/settings.json 2. CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING When set to "1," this disables adaptive thinking and reverts to the old fixed-budget system. Adaptive thinking was introduced with Opus 4.6 on February 5th. It dynamically allocates thinking tokens per turn: complex tasks get more, simple tasks get less. Critically, adaptive thinking on Opus 4.6 enables interleaved thinking, where the model reasons between tool calls, not just at the start of a response. This is confirmed in Anthropic's docs (platform.claude.com/docs/en/build-…). Disabling adaptive thinking on Opus 4.6 means you lose interleaved thinking entirely. So you are not unlocking deeper reasoning. You are reverting to a deprecated code path that gives you a fixed budget and removes a feature. 3. MAX_THINKING_TOKENS Sets the fixed thinking budget in tokens. Only takes effect when adaptive thinking is disabled. The default is 31,999. That is not a secret number. It is in the official docs (code.claude.com/docs/en/costs): "Extended thinking is enabled by default with a budget of 31,999 tokens." The number sits just under the 32K boundary where long-running requests can hit system timeouts. Setting MAX_THINKING_TOKENS to 31999 is setting it to what it already was. WHY SOME PEOPLE FEEL A DIFFERENCE: If your effort level silently dropped to "medium" after the March 4th update, and then you added DISABLE_ADAPTIVE_THINKING + MAX_THINKING_TOKENS=31999, you forced Claude back to thinking on every turn with a fixed budget. It would feel like "smart Claude is back" because you went from reduced thinking to always thinking. But the simpler fix was just setting effortLevel back to "high." WHEN DISABLING ADAPTIVE THINKING ACTUALLY HELPS: There is one documented case. Users with extensive CLAUDE.md files, custom skills, and strict coding rules reported that high-effort adaptive thinking sometimes deprioritizes those instructions (GitHub issue #23936, open, with multiple confirmations). Setting effort to "medium" or disabling adaptive thinking restored instruction-following for them. If that describes your setup, the nuclear option is valid: { "effortLevel": "high", "env": { "CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING": "1", "MAX_THINKING_TOKENS": "31999" } } But understand the tradeoff: you lose interleaved thinking on Opus 4.6, you lose dynamic allocation, and you cap every turn at 31,999 tokens even when a task could benefit from more. FOR MOST PEOPLE: { "effortLevel": "high" } One line. Keep adaptive thinking on. Type "ultrathink" in your prompt when you need maximum depth on a specific turn (re-introduced in v2.1.68). You get fast responses on simple tasks, deep reasoning when you ask for it, and interleaved thinking between tool calls. Read the actual docs before copying config from viral posts: Settings: code.claude.com/docs/en/settin… Adaptive Thinking: platform.claude.com/docs/en/build-… Costs: code.claude.com/docs/en/costs Changelog: github.com/anthropics/cla…



@levelsio Do you voice prompt or type? From the typos I see, I think you type. Have you tried voice?























