The oily silver rain of May 1, 2072, fell upon the Cascade Sprawl like a plague of locusts on a rotten frontier, bringing with it a novel prion variant that began to rewrite human biology. I tightened my grip on my lever-action rifle, a rusted relic of defiance in a cyberpunk wasteland, as I watched my wife, Sarah, pacing the floorboards of our decaying apartment. She had only gone out to scavenge for clean rations, but she returned profoundly changed, her skin shifting with iridescent metallic highlights, her movements possessing the fluid, predatory grace of a wild cougar, and her eyes catching the dim neon light like polished coins. The whispered legends spoke of a promised land beneath 47th Street called The Compound, a pre-collapse military bunker holding a pure water source and a localized cure, guarded by the ruthless Iron Veins outlaws and Sister Mara's crazed rain-worshipping zealots. My peripheral vision abruptly flared with ethereal blue text detailing my humanity index and newly crystallized skills like Pharmaceutical Looting and Spine-Aware Combat, while my hand instinctively brushed my meager inventory: a single dose of expired suppressant, a tarnished paper crane folded in Sarah's memory, and a crackling radio tuned to Captain Harlan's grim broadcast. The realization that my own countdown had started hit me with a crushing pressure behind my eyes, the unmistakable arrival of Stage One: The Headache. A copper tang flooded my mouth that no amount of canteen water could wash away, and as I stepped toward the glass to confront the shadowy, ashen reflection staring back at me from the rain-streaked window, a chilling blue notification flashed across my retinas: [System Alert: Foreign Prion Detected. Adapt or Perish?].
Kimi K2.6 — open-weight MoE with 256K context, text and image input. Sits between gpt-5.4-mini and gpt-5-mini on price with far more context.続きを読む →
Kimi K2.7 Code2
速度
Kimi K2.7 Code — agentic coding model, 1T-parameter MoE with 256K context. Handles text, image and video input; tuned for long-context programming tasks.続きを読む →
Qwen Max2
速度
Alibaba Qwen Max — フラッグシップの推論モデル。複雑なタスク、深い分析、構造化された出力でトップクラスの性能を発揮します。32K トークンのコンテキスト。続きを読む →
Qwen VL Max2
速度
Alibaba Qwen VL Max — マルチモーダルのフラッグシップ。テキストとともに画像を入力でき、テキストを返します。ビジュアル Q&A、OCR、画像に基づく推論に利用できます。続きを読む →
Gemini 3.5 Flash — 優れた検索とグラウンディングを備え、速度に最適化された Google の最先端知能モデル。エージェントループ、長期的なコーディング、迅速な反復作業に優れています。1M トークンのコンテキスト。続きを読む →
Gemini 3.7 Flash3
速度
Gemini 3.7 Flash — Google's fast frontier model with upgraded reasoning for software engineering, web development and agentic workflows. Successor to Gemini 3.5 Flash. 1M-token context.続きを読む →
Kimi K2.7 Code HighSpeed3
速度
Kimi K2.7 Code HighSpeed — same model as K2.7 Code at roughly 180 tokens/sec output, for exactly 2x the price. Pick this only when latency matters more than cost.続きを読む →
Kimi K3 — Moonshot's flagship, 2.8T-parameter MoE with native visual understanding and a 1M-token context. Priced level with Claude Sonnet 4.5 for materially more context.続きを読む →