TangZhiYuan

🚀 MiniMax-M3 is now live on RoundRouter! Multimodal, 1M-token context, and built for the long game. ⚡ One API gets you OpenAI, Claude, Gemini, and MiniMax, with the speed and stability builders actually need. 🔥 Global, secure, compliant access, and every model is 50% off right now. roundrouter.com/sign-up?aff=BP…

🚀 Qwen3.7-Plus is now live on RoundRouter. 👀 One model that can see, reason, code, and act — built for multimodal workflows across text, vision, GUI, and CLI. ⚡ From coding copilot to productivity assistant to visual agent, Qwen3.7-Plus is made for people who want one model that actually gets things done. 🌍 With RoundRouter, you don’t need to juggle multiple accounts or complicated setups. 🔑 One API account gives you direct access to native AI services from OpenAI, Claude, Gemini, Qwen, and more. 🛡️ Built for stability, speed, and secure access to top-tier AI across different network environments. 🔥 And right now, all model usage is 50% off. 💡 If you’ve been looking for a cleaner, faster way to build with the best models in one place, this is a good day to start. 👉 Try Qwen3.7-Plus on RoundRouter and build something impressive. roundrouter.com/sign-up?aff=BP…

It might be an exaggeration to say that google has fallen out of the top three, but the underlying implication that it has failed to translate its massive lead into dominance is certainly true. It’s not that google isn’t strong; it’s just that it falls far short of the level of dominance people originally expected.

google fell off. harvested 20 years of humanity’s data and still can’t crack the top 3 in the AI race. smh.

@aleabitoreddit 真正有认知差的机会，本来就不在最热闹的概念层，而是在产业链里那些不起眼但卡住全局的瓶颈节点。InP这种东西，平时没人看，等市场一致性认知形成时，往往已经不是最舒服的位置了。

中文社区又出了一篇佳作！ 当我发现像磷化铟（InP）衬底这样的产业链瓶颈或关键节点时，我通常会凭主观判断来进行操作。 这在很大程度上取决于能否准确预判下一次的技术架构演进！ 尤其是去年，当所有人都在说光技术还为时过早、铜互连不会被取代时，我果断做多了光通信供应链。 看到大家如何解读我的投资思路，并将其转化为更偏算法化、系统化的逻辑，真的非常有意思。

天涯论坛真的太精彩了！10年前的天涯社区卧虎藏龙，他们一部分人来自京城大院、并且这时候天涯里充斥着高人、各种思想碰撞使得许多神贴应运而生、各种神帖让人看得停不下来。 真怀念当年在天涯自由追帖的日子，那纯粹的年代一去不复返咯。

@heynavtoor Extra tip: If you need to change your model in the future, you might want to consider using RoundRouter, which allows you to configure everything in one place. Currently, all models are available at 50% OFF.

Everyone knows ChatGPT, Claude, Grok, and Gemini. Here are 10 underdog AI tools the smart 1% are quietly using in 2026. 1. DeepSeek Free GPT-4-class reasoning at zero cost. The Chinese AI that shook the entire industry by matching ChatGPT for 1/10th the training cost. The rest of the world is still pretending it doesn't exist. Site → chat.deepseek.com 2. Kimi Free 2 million token context window. Longer than any Claude, ChatGPT, or Gemini model. Drop in entire books, full codebases, or 100 PDFs at once. Site → kimi.ai 3. Qwen Alibaba's flagship AI. Ranks top three on global benchmarks against ChatGPT and Claude. Free, multilingual, aggressive on coding tasks. Site → chat.qwen.ai 4. Mistral Le Chat The fastest AI responses on the market. French-built, multilingual, free. Mistral is the European answer to the US AI giants. Site → chat.mistral.ai 5. You AI search with built-in image generation, code execution, and citations. The Swiss Army knife of AI search engines. Site → you.com 6. Komo Search Privacy-first AI search. No tracking, no ads, no data harvesting. The DuckDuckGo of AI. Site → komo.ai 7. Krea AI Real-time AI image generation that updates as you type. Watch images form in front of you. Free tier ships hundreds of images. Site → krea.ai 8. Glasp AI highlights and summaries for any article or YouTube video. The reading assistant every serious researcher quietly uses to read 10x more. Site → glasp.co 9. HuggingChat Free chat with 50+ open-source AI models in one interface. The playground every AI engineer uses to compare models before paying for one. Site → huggingface.co/chat 10. Lepton AI Free fast inference for any open-source AI model. Run Llama, Mistral, and Qwen at GPT-4 speeds without owning a GPU. Site → lepton.run

⚡CC Switch v3.16.0 is a big Update. Codex can now connect to third-party Chat Completions models with almost no friction: DeepSeek, Kimi, GLM, SiliconFlow, MiniMax, and more. The best part: it doesn’t feel like a hacky workaround. Reasoning blocks, tool calls, chat history, and streaming all carry over cleanly through the local proxy layer. For most users, setup is basically a 10 minute job: 1️⃣ Download and install the latest version GitHub Releases: github.com/farion1231/cc-… Download the version for your system (Mac: .dmg recommended; Windows: .msi recommended) Install and open it; on first launch, you can import old settings with one click. 2️⃣ Switch to the Codex panel Click the Codex icon at the top → Enter the Codex management interface. 3️⃣ Add third-party providers (most critical) Click the + Add button in the top-right corner Select a preset (DeepSeek, Kimi, GLM, SiliconFlow, MiniMax, etc.; v3.16 adds a large number of Codex-specific presets) Fill in: • Provider name • API Key (obtain from the corresponding platform) • Base URL (the preset automatically includes /v1) Check/confirm to enable the Chat Completions to Responses routing (key new feature) Save → Restart Codex (or restart your IDE/terminal) for the changes to take effect. 4️⃣ Testing Type `codex` in the terminal or invoke it in VS Code, then ask a question. If you see a normal reasoning process and an answer, you’ve succeeded! You can also quickly switch providers via the system tray. Common Tips • Recommended combinations: DeepSeek (best value), Kimi (long text), SiliconFlow (stable) • Session history is isolated by provider; simply create a new session after switching (this is not a bug) • Supports usage display, unified MCP management, and Skills synchronization • Switch back to the official version anytime Resources • GitHub: github.com/farion1231/cc-… • Official Release Notes: docs/release-notes/v3.16.0 in the repository • Author: @Jason_Young1231

🚀 Nvidia, Windows, and Arm posting together? That’s not a tease: that’s a signal. 💻 A new PC era might be closer than people think.

看到别人晒出收益有种感慨，这样的大环境下，这个平台不要太友善。 #蓝V互助

@8mXJvt4ZidmemYc 大家多多关注一下我这个新人，甚是感谢！

娯楽であるはずのTwitterで130万円以上稼げるの、どう考えてもバグでしょ…

@Henri12977 财政供养率很高

Le marché du travail est une blague en France. Il n’y a plus d’emplois nulle part, excepté pour des boulots de bas étage payés au SMIC. Seuls 30 % de la population travaillent dans le secteur privé, le reste étant des fonctionnaires, des chômeurs ou des retraités. Les jeunes n’ont jamais eu si peu d’opportunités professionnelles.

@indrawxyz Extra tip: If you need to change your model in the future, you might want to consider using RoundRouter, which allows you to configure everything in one place. Currently, all models are available at 50% OFF.

Wah gila opus 4.8 baru rilis tapi tokennya udah dijual harga diskon 80% di website china wkwk🔥 tinggal dipilih sesuai kebutuhan, ada pilihan group claw/hermes kalau mau dipakai buat ai agent. bisa topup pakai crypto, minimal 50 yuan (~130 ribu rupiah).

@elonmusk @JeffBezos @zerohedge 都是伟大的人

@JeffBezos @zerohedge Ad astra per aspera

All personnel are accounted for and safe. It’s too early to know the root cause but we’re already working to find it. Very rough day, but we’ll rebuild whatever needs rebuilding and get back to flying. It’s worth it.

现在App自动续费在2026年还能坑人的原因也就呼之欲出了 老年人不熟悉免费试用+自动续费的弯弯绕绕，而苹果又乐于看到开发者采用订阅模式 这样的信息差就给了无良开发者钻空子的机会。

@JapanTank In the past, we were looking for tools; now, it’s more like we’re building them ourselves. If you frequently switch between different models, having a unified configuration is actually quite important, and RoundRouter’s approach is a good one.

Claude CodeでもCodexでもAntigravityでも、「YouTubeの動画デスクトップにダウンロードするプログラム作って」と言えば数分で作るんだもんなあ。 IT知識のない小学生でもできてしまう。 欲しいツールはその場で作るのが当たり前になった。

@boxmining @claudeai Extra tip: If you need to change your model in the future, you might want to consider using RoundRouter, which allows you to configure everything in one place. Currently, all models are available at 50% OFF.

Got banned from @claudeai while testing Opus 4.8 because I burned through 80% of my token limit in 30 minutes running agent teams. Is that a flex or a cry for help?

新的Opus4.8 vs Opus4.7 分别模拟3个真实的物理场景表现如何？

@CEO_loves_tech Just a tip: If you want more flexibility for future model changes, RoundRouter lets you configure everything in one place. Right now, all models come with 50% OFF.

いい感じのVPSを借りて、CodexとClaude CodeとHermes Agentを入れて、MacbookやiPhoneから24/365で指示しまくる環境が一番満足度高いと思ってる。 ローカルLLMを入れたいとしても、せいぜい月1万円くらいのインフラ代（LLM利用料除く）で遊び倒せる。 皆なんでMacMiniとか買うのさ。

@Ice_Hazymoon Claude：互相蒸馏是吧？看我整波大的

用 Qwen 训练后的 Claude

@jeantuite 松弛，真羡慕

Conselho do dia: Não se estresse, a empresa não é sua e hoje é sexta-feira.