Fresh stories
Tests link Ox Alpha to Zhipu GLM API routes and error codes
Researchers say malformed Ox Alpha requests exposed Zhipu-specific routes, error codes, and an internal class name. Independent vision comparisons also argue against speculation that the stealth model is Gemini.
Tests link Ox Alpha to Zhipu GLM API routes and error codes
Researchers say malformed Ox Alpha requests exposed Zhipu-specific routes, error codes, and an internal class name. Independent vision comparisons also argue against speculation that the stealth model is Gemini.
Anthropic says serving test remapped Claude Code effort settings
Anthropic says a test serving configuration mapped Claude Code’s numeric effort settings differently, allowing “high” to display as 10. The company says evaluations found no regression and the underlying models were unchanged.
Qwen 3.8 27B reaches 3,200 TPM at 262K context on two RTX 3090s
Community tests report Qwen 3.8 27B handling coding, OCR, and long-context workloads locally. One vLLM setup reached 3,200 tokens per minute at 262K context on two RTX 3090s without NVLink.
Top storiesthis week
OpenAI grants Codex customers a banked usage reset
OpenAI gave paid ChatGPT Work and Codex users a banked usage reset and said Codex has reached 20 million active users. The company is investigating reports that lower cache-hit rates are causing usage limits to drain faster.
multiPL-E regex bug corrupts MBPP benchmark language variants
An audit found that multiPL-E's MBPP subset replaced every occurrence of "py" rather than the word "python," creating malformed language names. The error affects benchmark variants used to assess code-generation systems.
OpenAI cuts GPT-5.6 Sol API rates to $4 input and $20 output
OpenAI cut GPT-5.6 Sol API input and output rates from $5 and $30 to $4 and $20 per million tokens for three months. Subscription usage remains unchanged.
Claude Security adds Mythos 5 scans for GitHub repositories
Claude Security's public beta now uses Mythos 5 to scan GitHub repositories for enterprise customers. Findings include CWE labels, severity, confidence, and suggested patches that can open in Claude Code on the web.
Study finds CLI-first agents cost 5–28x less than MCP agents
A study across seven agents and five models found CLI-first agents were as reliable on mature software tasks as MCP-enabled agents. The CLI setups cost 5–28 times less in the reported experiments.