AI teams often hit a 'latent capability ceiling' where larger models no longer improve task accuracy, with scaling data showing diminishing returns beyond 100B parameters and 61% of tasks exhibiting n…
A new study from Japan's AI Security Institute finds that LLM-as-judge safety evaluations are not reproducible even at temperature 0, with per-item disagreement up to 50% across runs. The researchers …
Tsvetan Stoychev, a former principal software engineer at Akamai Technologies, used AI tools including GitHub Copilot, Claude Opus, and Gemini to find vulnerabilities in the ClickHouse codebase, succe…
GitHub evaluated its Copilot agentic harness across multiple benchmarks and models, finding it achieves task completion rates on par with vendor harnesses while using fewer tokens. The harness powers …
Zhipu AI released GLM 5.2, a 744B parameter open-weight mixture-of-experts model that rivals Claude Opus in coding and visual design quality at lower inference cost. The model's MoE architecture activ…
A developer reports that AI coding costs can be slashed by 70% through task-level model routing instead of using a single frontier model for all tasks. After profiling usage, the developer found that …
Z.ai benchmarked GLM-5.2 against Claude Opus on 45 coding tasks using an identical agent setup, finding both models solved 25 tasks and agreed on 43 of 45 outcomes. With prompt caching enabled, GLM-5.…
A developer identifies five common ways AI coding agents waste tokens, including prompt caching inefficiencies, long-context bloat, oversized tool schemas from MCP servers, and repeated context. The p…
Zhipu AI released GLM 5.2, a 744-billion-parameter mixture-of-experts open-weight model that outperforms Claude Opus on visual design tasks at lower cost. The model features a longer context window, i…
Z.ai's GLM-5.2, a 756B-parameter model with a 1M-token context window, competed against Claude Opus in a Tech Stackups test building a 3D platformer in raw WebGL. Claude Opus finished in 33m 30s at an…
An engineer at LiteLLM reveals five cost traps in AI API gateways that can silently inflate bills, including retry logic that multiplies API calls across fallback chains and misconfigured fallbacks th…
Zhipu AI's open-weight GLM 5.2 model scored 39% F1 on IDOR detection in Semgrep's benchmarks, outperforming Claude Code (32%) at $0.17 per vulnerability. The model, released June 13, 2026, with MIT-li…
A team of 40 engineers using Claude Code coding agents saw a 340% increase in AI costs, reaching $20K/month in unexpected spend due to raw API keys without per-developer budgets or team caps. The team…
A developer used Claude Opus to generate a mid-century country-club-core todo-list app in Python, requiring five essays on obscure aesthetics and programming paradigms before any code was written. The…
A developer reports that Claude Opus repeatedly refactors their novel codebase due to pattern recognition from training data, causing weeks of manual fixes. The user claims their work involves cutting…
A developer built an autonomous incident investigation agent called FRIDAY that reduced mean time to resolution by 65% on a platform serving 30+ million end users. The agent, triggered by PagerDuty al…
A software founder details how local Qwen models have provided real value for his business, paying for a GPU within months, but warns of infinite loops and hallucination risks, especially when quantiz…
MiniMax released the M3 open-weight model, claiming it costs 5% of Claude Opus per task, achieves 59% on SWE-Bench Pro, and supports a 1-million-token context window at one-twentieth the compute of it…
Z.ai released GLM-5.2, an open-weights AI model under an MIT license, positioning it between Claude Opus 4.7 and 4.8 in performance while costing less than a fifth of Opus on output tokens. The model …
A team of researchers and organizations launched A Beginner’s Guide to Digital Minds, a website introducing AI consciousness, AI welfare, and the moral implications of digital minds. The guide offers …