Xiaomi MiMo-V2.5-Pro Rivals Claude Opus Coding
Xiaomi's new open-weight model nearly matches Claude Opus 4.6 on coding benchmarks while using 40-60% fewer tokens.
📂 LLM News
Latest articles in LLM News
Same Prompt, Different Morals: AI Models Split on Ethics
A new benchmark testing 100 ethical scenarios reveals stark divergence among leading AI models on moral reasoning.
DeepSeek V4 Flash Faces Early Criticism Over Quality
Early users report DeepSeek V4 Flash underperforms rivals like Qwen 3.6 Plus and GLM-5 in instruction following and long…
DeepSeek V4 Flash Faces User Backlash Over Real-World Performance
Heavy users report DeepSeek V4 Flash underperforms rivals like Qwen 3.6 Plus and GLM5 in instruction following and long-…
DeepSeek V4 Flash Draws Mixed Reviews From Power Users
Early adopters report DeepSeek V4 Flash underperforms rivals in instruction following and long-context tasks despite the…
DeepSeek V4 Benchmarks Show 8-Month Gap With US AI Leaders
DeepSeek's latest V4 model series arrives 15 months after R1, but benchmark comparisons suggest it still trails top US m…
The 'Chinese Tax' on AI Tokens: Why Non-English Costs More
Chinese text consumes up to 2x more tokens than English in most LLMs. Here's why tokenizers create an invisible language…
Grok 4.3 Quietly Launches With Mixed Benchmark Results
xAI releases Grok 4.3 as a pragmatic upgrade — cheaper and faster, but still trailing GPT-5.5 and Claude Opus 4.7 in key…
Third-Party GPT-5.5 API Relay Services Emerge With Free Credits
A growing ecosystem of API relay services now offers shared access to OpenAI's latest GPT-5.5 Pro accounts, raising ques…
The Memory Illusion: Why Your LLM Doesn't Actually Remember You
LLMs feel like they remember your conversations, but architecturally they start from scratch every time. Here is how the…
Batch or Split? How to Send Multiple Questions to LLMs Faster
When you have multiple unrelated questions for an LLM, splitting them into parallel requests almost always beats batchin…
Claude Now Has Three Memory Systems — Here Is Why
Anthropic ships three distinct memory architectures across Claude.ai, Claude Code, and the API, each with different trad…