Tagged articles : 언어 모델 (LLM)

98.6% on the hardest test, 61 out of 100 from the independent judge: where the gap comes from, and what GPT-6 Astra really changes.

Translate in both directions with a single key, in 28 languages, using OpenAI, Claude, Gemini, Grok or Ollama. Windows, macOS and Linux, x64 and...

Nvidia pays $6 billion for the machine that makes Poolside's models. We open the hood: the mixer, the furnace, the tests and the finishing touches.

A chip that does just one thing, one machine delivered, and a customer that invested $700 million after testing it for a month.

Corrected comparison of AI coding tools: what to choose for your profile, price per million tokens, and actual limits.

Astra has several agents work together for days. Ten math problems for $2,000, and development slowed over cybersecurity concerns.

Meta releases Muse Glimmer under Apache 2.0: 30 billion parameters, under 20 GB, and 233 tokens/s on an RTX 5090. See how it compares with Opus 4.8.

ByteDance is training a model with up to 10,000 billion parameters. Why this figure says almost nothing, and what the Doubao story really hides.

DeepSeek reports 82.7 on Terminal-Bench versus Opus 4.8’s 85.0. The official leaderboard says 78.9. What the evaluation harness really changes.