3 posts
MMLU, HumanEval, SWE-bench Verified/Pro, ARC-AGI-2, GPQA Diamond, AIME, LiveCodeBench v6, Terminal-Bench 2.0, OSWorld, HLE, plus Turkish benchmarks (TR-MMLU, TUMLU) — what each one measures, the frontier thresholds, contamination and cherry-picking risks, and practical meaning for CTOs, investors, and engineers. 32+ references.
A head-to-head comparison of the two 2026 flagship AI models — Anthropic Claude Opus 4.7 and OpenAI GPT-5. Architecture and training philosophy differences (Constitutional AI vs RLHF), benchmark results (MMLU, HumanEval, GSM8K, hallucination), Turkish performance, code generation, reasoning, long context (1M vs 256K), multimodal, agent/tool use/MCP, cost, latency, safety, and alignment. Use-case-based winner analysis.
The most comprehensive 2026 Turkish LLM benchmark: MMLU-TR, Belebele-TR, TruthfulQA-TR, Turkish HumanEval, MGSM-TR, and hallucination tests. Score tables for GPT-5, Claude Opus 4.7, Gemini 3, Mistral Large 3, Llama 4, DeepSeek V3, Qwen 2.5, and local Turkish models (Cezeri, BERTurk, Trendyol-LLM), with use-case mapping and transparent methodology.