Llm on Chirag Hasija

Llm on Chirag Hasija https://chiraghasija.cc/tags/llm/ Recent content in Llm on Chirag Hasija Chirag Hasija https://chiraghasija.cc/og-image.png https://chiraghasija.cc/og-image.png Hugo -- 0.155.3 en-us Thu, 05 Feb 2026 16:09:00 +0530 The Open Source Model That Beat GPT-4o at Half the Cost https://chiraghasija.cc/posts/open-source-model-beat-gpt4o/ Thu, 05 Feb 2026 16:09:00 +0530 https://chiraghasija.cc/posts/open-source-model-beat-gpt4o/ Open-weight models like Llama 4, Mistral, and Qwen have reached GPT-4o performance on many tasks at significantly lower cost. This post covers when self-hosted open models make sense. Claude vs GPT-4o vs Gemini: A Benchmark That Actually Matters https://chiraghasija.cc/posts/claude-gpt4o-gemini-benchmark-that-matters/ Wed, 26 Nov 2025 17:38:00 +0530 https://chiraghasija.cc/posts/claude-gpt4o-gemini-benchmark-that-matters/ A practical comparison of Claude, GPT-4o, and Gemini based on real engineering tasks - code generation, debugging, instruction following, and reasoning - not academic benchmarks. The Real Cost of Running LLMs in Production https://chiraghasija.cc/posts/real-cost-running-llms-production/ Sun, 17 Aug 2025 19:03:00 +0530 https://chiraghasija.cc/posts/real-cost-running-llms-production/ The true cost of running LLMs in production goes far beyond per-token pricing. This post breaks down infrastructure, latency, and hidden costs that most teams miss. How Meta Trains LLaMA 4 on 100,000 GPUs https://chiraghasija.cc/posts/how-meta-trains-llama4-100k-gpus/ Sat, 09 Aug 2025 09:42:00 +0530 https://chiraghasija.cc/posts/how-meta-trains-llama4-100k-gpus/ An exploration of the distributed systems and hardware engineering required to train LLaMA 4 on tens of thousands of GPUs, and what limits performance at this scale.