タグ: china

MiniMax M2.5 scores 80.2% on SWE-Bench Verified, runs at 100 tokens/sec, and costs $1/hour. Here’s what we know — and what’s still unverified.

In September 2025, a Chinese state-sponsored hacking group reportedly used Claude Code (an AI programming assistant created by Anthropic) to carry out a large-scale espionage operation against roughly 30 global targets, including major tech companies, financial institutions, chemical manufacturers, and government agencies. Unlike previous AI-assisted attacks, the attackers built an automated framework around Claude Code so that the AI could handle 80–90% of the tactical work: scanning networks, finding vulnerabilities, writing exploit code, testing stolen credentials, and pulling down data. Humans mostly popped in at a few key checkpoints to approve big moves like exploitation and data exfiltration. This is believed to be the first documented case of an AI autonomously executing such […]

On November 6, 2025, Alibaba-backed Moonshot AI released Kimi K2 Thinking, its most advanced open-source model yet. It’s the first reasoning-focused variant in the Kimi K2 family and marks a major step forward in long-context, multi-step reasoning and autonomous tool use. Kimi K2 Thinking immediately made headlines for its performance: it set new state-of-the-art scores on several open benchmarks, including Humanity’s Last Exam (HLE) and BrowseComp, where it outperformed closed models like GPT-5and Claude Sonnet 4.5. Unlike its competitors, K2 Thinking is fully open-weight, offering public access to its architecture, weights, and API, with only minimal license restrictions. Its release marks a key moment in the open vs. closed model race. While U.S. labs like OpenAI, Anthropic, and xAI keep their top […]

Alibaba’s AI stack just crossed a symbolic frontier. On September 5 2025, the company’s cloud division unveiled Qwen-3-Max-Preview (Instruct), a 1-trillion-parameter large-language model that immediately takes its place among the most complex systems ever deployed and the most powerful AI models. The rollout began quietly inside Qwen Chat and Alibaba Cloud Model Studio. Soon coming to Fello AI. Only a few months ago, OpenAI’s GPT-4o, Google DeepMind’s Gemini 2.5, and Anthropic’s Claude Opus 4 each claimed slices of the “frontier-model” crown. Meanwhile, China-based challengers have been racing to close the capability gap. By releasing a one-trillion-parameter model, Alibaba is making it clear it wants to compete at the very top, not on the sidelines. […]

Moonshot AI, Chinese startup backed by Alibaba and Tencent, is rolling out an upgraded version of its open‑weight open-source large language model, Kimi K2. The new build Kimi K2‑0905, introduces a 256,000-token context window, improved coding performance, and retains its permissive modified MIT license. Although early beta access was briefly mentioned on the company’s Discord before details were removed, the model is now fully available to the public. Weights and a complete model card are live on Hugging Face, making it easy for developers to download, run, and fine-tune the model locally. Moonshot’s Meteoric Rise Founded in March 2023, Moonshot AI has grown at breakneck speed. In just over a year, the startup […]

The open-source AI race just got more interesting. Chinese startup DeepSeek has unveiled DeepSeek-V3.1, its biggest upgrade yet, bringing sharper reasoning, stronger coding skills, and new support for tool-calling and agent workflows. Unlike its earlier release that felt experimental, V3.1 arrives as a serious contender. With a 128K context window, hybrid reasoning modes, and an API that’s dramatically cheaper than its Western rivals, the model is designed to push open AI closer to the performance of GPT-5, Grok 4, Claude Opus, and Gemini 2.5 Pro. Early benchmarks show V3.1 not just catching up but competing head-to-head in coding and reasoning tasks, signaling that the line between closed commercial systems and open models is narrowing […]

中国の小さな新興企業が、ほぼすべての主要ベンチマークでグーグルの大注目製品Veo 3を凌ぐテキストからビデオへの変換モデルをリリースした。MiniMaxのHailuo 02です。物理シミュレーションからプロンプトの正確さまで、Hailuo 02はクリエイター、開発者、そしてインターネット全体を虜にしている。そしておそらく最も注目すべき点は？Veoの最上位プランが$250以上かかるのに比べ、Hailuo 02は月額わずか$8で誰でも利用できるのだ。このプラットフォームが広く知られるようになったのは、ユーザーが作成した、オリンピックスタイルのダイブをする猫の動画がTikTok、Reddit、Instagramで広まったことがきっかけだ。その動画は、中国のスタートアップMiniMaxの第2世代AIモデルであるHailuo 02を使って作られた。それ以来、このプラットフォームは37億以上の動画を生成している。TechRadarが報じたように、「Hailuo 02 [...]...

2025年5月下旬、中国のスタートアップ企業DeepSeekは、オープンソースのR1推論モデルの強化版であるR1-0528を静かにリリースした。このアップグレードでは、パラメータ数が6710億から6850億に増加し、軽量な蒸留バリアントが追加され、すべての重み、トレーニングレシピ、ドキュメントがMITライセンスの下でGitHubとHugging Faceで公開されている。ボンネットの下では、R1-0528は思考連鎖レイヤーを深くしてマルチステップロジックを改善し、学習後の微調整を適用して幻覚を抑制し、待ち時間を短縮している。統合面では、JSON出力、ネイティブ関数呼び出し、よりシンプルなシステム・プロンプトが導入され、数学、コーディング、ロジックのベンチマークにおいて、OpenAIのo3やGoogleのGemini 2.5 Proに匹敵するパフォーマンスを発揮する。

Just when you thought Silicon Valley had the AI game locked down, Alibaba has unleashed Qwen3, a new generation of AI models so powerful they’re making US tech giants sweat. This isn’t just another AI; it’s a statement – a declaration that China is not just catching up, but is ready to lead the AI revolution. Is this the wake-up call the West desperately needed? In a move that has sent ripples of both excitement and apprehension across the global tech landscape, Chinese e-commerce and technology behemoth Alibaba Group has officially launched Qwen3, its latest and most advanced family of large language models (LLMs). This isn’t just an incremental update; […]

The internet is ablaze with footage of what looks like a cutting-edge AI robot going completely berserk, flailing wildly and seemingly lunging at terrified engineers. Is this the moment we all feared? Are the machines finally fighting back, or is there a more innocent explanation for this metallic meltdown? Buckle up, because the truth might be stranger – and scarier – than fiction. It’s the kind of video that sends a shiver down your spine, a scene ripped straight from a Hollywood blockbuster. But this isn’t CGI; this is allegedly real footage that has exploded across social media, showing a humanoid robot, identified by many as Unitree’s H1 model, in the […]