LLM on Osmond van Hemerthttps://www.osmondvanhemert.nl/tags/llm/Recent content in LLM on Osmond van HemertHugo -- gohugo.ioen© Osmond van Hemert. All rights reserved.Thu, 29 Jan 2026 00:00:00 +0000AI Agent Frameworks — The Wild West of Autonomous Systemshttps://www.osmondvanhemert.nl/posts/260129-ai-agent-frameworks-landscape/Thu, 29 Jan 2026 00:00:00 +0000https://www.osmondvanhemert.nl/posts/260129-ai-agent-frameworks-landscape/The AI agent framework landscape has exploded, with LangGraph, CrewAI, AutoGen, and dozens more competing for developer mindshare. Here’s what matters.Google Gemini 2.0 — A New Chapter in Multimodal AIhttps://www.osmondvanhemert.nl/posts/251211-google-gemini-2-multimodal-ai/Thu, 11 Dec 2025 00:00:00 +0000https://www.osmondvanhemert.nl/posts/251211-google-gemini-2-multimodal-ai/Google launches Gemini 2.0 with native multimodal capabilities, and the implications for developers are significant.SWE-bench Benchmark Contamination — When the Test Answers Are in the Training Datahttps://www.osmondvanhemert.nl/posts/250911-swe-bench-git-history-leaks/Thu, 11 Sep 2025 00:00:00 +0000https://www.osmondvanhemert.nl/posts/250911-swe-bench-git-history-leaks/Research reveals that top AI coding model scores on SWE-bench may be inflated due to git history leaks, raising fundamental questions about how we evaluate AI coding capabilities.Mistral's Le Chat Gets MCP Connectors — The Protocol That's Quietly Connecting Everythinghttps://www.osmondvanhemert.nl/posts/250904-mistral-le-chat-mcp-connectors/Thu, 04 Sep 2025 00:00:00 +0000https://www.osmondvanhemert.nl/posts/250904-mistral-le-chat-mcp-connectors/Mistral adds custom MCP connectors and persistent memory to Le Chat, signaling that the Model Context Protocol is becoming the standard glue for AI tool integration.Google's Gemma 3 270M — Why Tiny Models Are the Real AI Storyhttps://www.osmondvanhemert.nl/posts/250814-gemma3-270m-small-models-big-impact/Thu, 14 Aug 2025 00:00:00 +0000https://www.osmondvanhemert.nl/posts/250814-gemma3-270m-small-models-big-impact/Google releases Gemma 3 at 270M parameters, proving that smaller, more efficient models might matter more than the next big model launch.GPT-5 Is Here — A Developer's First Look at What Actually Changedhttps://www.osmondvanhemert.nl/posts/250807-gpt5-launch-developer-implications/Thu, 07 Aug 2025 00:00:00 +0000https://www.osmondvanhemert.nl/posts/250807-gpt5-launch-developer-implications/OpenAI launches GPT-5 with significant improvements. Here’s what matters for developers beyond the marketing.The EU AI Act Compliance Clock Is Ticking — What Developers Need to Knowhttps://www.osmondvanhemert.nl/posts/250703-eu-ai-act-developer-compliance/Thu, 03 Jul 2025 00:00:00 +0000https://www.osmondvanhemert.nl/posts/250703-eu-ai-act-developer-compliance/With key EU AI Act provisions now in effect, development teams building AI systems need to understand the practical implications for their architectures and workflows.OpenAI's o3 and o4-mini — Reasoning Models Get Realhttps://www.osmondvanhemert.nl/posts/250417-openai-o3-o4-mini-reasoning-models/Thu, 17 Apr 2025 00:00:00 +0000https://www.osmondvanhemert.nl/posts/250417-openai-o3-o4-mini-reasoning-models/OpenAI releases o3 and o4-mini reasoning models, bringing chain-of-thought inference to mainstream developer workflows.Claude 3.7 Sonnet — Extended Thinking Changes the Game for AI-Assisted Developmenthttps://www.osmondvanhemert.nl/posts/250306-claude-3-7-sonnet-extended-thinking/Thu, 06 Mar 2025 00:00:00 +0000https://www.osmondvanhemert.nl/posts/250306-claude-3-7-sonnet-extended-thinking/Anthropic’s Claude 3.7 Sonnet introduces extended thinking, letting the model reason step-by-step before responding — and the implications for developer workflows are significant.Claude 3.5 Gets a Computer — Anthropic's 'Computer Use' and the Future of AI Agentshttps://www.osmondvanhemert.nl/posts/250220-anthropic-computer-use-ai-agents/Thu, 20 Feb 2025 00:00:00 +0000https://www.osmondvanhemert.nl/posts/250220-anthropic-computer-use-ai-agents/Anthropic’s computer use capability lets Claude interact with desktop applications like a human. What does this mean for automation, testing, and the future of AI agents?DeepSeek R1 — Open-Source Reasoning Models Change the Gamehttps://www.osmondvanhemert.nl/posts/250123-deepseek-r1-open-source-reasoning/Thu, 23 Jan 2025 00:00:00 +0000https://www.osmondvanhemert.nl/posts/250123-deepseek-r1-open-source-reasoning/DeepSeek’s R1 reasoning model, released as fully open-source with an MIT license, demonstrates that frontier AI capabilities aren’t exclusive to US labs anymore.Google Launches Gemini 2.0 Flash — The Multi-Modal AI Race Accelerateshttps://www.osmondvanhemert.nl/posts/241212-google-gemini-2-flash/Thu, 12 Dec 2024 00:00:00 +0000https://www.osmondvanhemert.nl/posts/241212-google-gemini-2-flash/Google’s Gemini 2.0 Flash brings native tool use, multimodal output, and agentic capabilities. A look at what this means for the competitive AI landscape.OpenAI Launches o1 Full Model and $200/Month ChatGPT Pro — The Reasoning Era Beginshttps://www.osmondvanhemert.nl/posts/241205-openai-o1-full-model-chatgpt-pro/Thu, 05 Dec 2024 00:00:00 +0000https://www.osmondvanhemert.nl/posts/241205-openai-o1-full-model-chatgpt-pro/OpenAI kicks off its ‘12 Days of OpenAI’ event with the full o1 reasoning model and a new $200/month ChatGPT Pro tier. What this means for developers building with AI.Claude Gets Hands — Anthropic's Computer Use Changes the AI Gamehttps://www.osmondvanhemert.nl/posts/241024-anthropic-claude-computer-use/Thu, 24 Oct 2024 00:00:00 +0000https://www.osmondvanhemert.nl/posts/241024-anthropic-claude-computer-use/Anthropic’s updated Claude 3.5 Sonnet introduces Computer Use, letting AI directly interact with desktop environments — a significant leap toward autonomous AI agents.OpenAI o1 — The Dawn of Reasoning Modelshttps://www.osmondvanhemert.nl/posts/240912-openai-o1-reasoning-models/Thu, 12 Sep 2024 00:00:00 +0000https://www.osmondvanhemert.nl/posts/240912-openai-o1-reasoning-models/OpenAI releases o1, a model that ’thinks before it answers’ — what chain-of-thought reasoning means for developers and the future of AI-assisted coding.GitHub Models — Bringing AI Model Experimentation to Where Developers Already Livehttps://www.osmondvanhemert.nl/posts/240822-github-models-ai-marketplace/Thu, 22 Aug 2024 00:00:00 +0000https://www.osmondvanhemert.nl/posts/240822-github-models-ai-marketplace/GitHub launches Models, a new playground for experimenting with AI models directly from GitHub. Here’s why this integration matters.Llama 3.1 405B — Meta Goes All-In on Open-Source AIhttps://www.osmondvanhemert.nl/posts/240725-meta-llama-3-1-open-source/Thu, 25 Jul 2024 00:00:00 +0000https://www.osmondvanhemert.nl/posts/240725-meta-llama-3-1-open-source/Meta releases Llama 3.1 with a 405 billion parameter model under a permissive license, making frontier-class AI genuinely open for the first time.Ollama and the Rise of Local LLMs — Why Running AI on Your Own Hardware Mattershttps://www.osmondvanhemert.nl/posts/240711-ollama-local-llm-revolution/Thu, 11 Jul 2024 00:00:00 +0000https://www.osmondvanhemert.nl/posts/240711-ollama-local-llm-revolution/Local LLM tooling has matured rapidly, with Ollama leading the charge. Here’s why self-hosted AI is becoming a serious option for developers.Claude 3.5 Sonnet — Anthropic Raises the Bar for Coding AIhttps://www.osmondvanhemert.nl/posts/240620-claude-35-sonnet-raises-the-bar/Thu, 20 Jun 2024 00:00:00 +0000https://www.osmondvanhemert.nl/posts/240620-claude-35-sonnet-raises-the-bar/Anthropic releases Claude 3.5 Sonnet, which benchmarks above GPT-4o on coding tasks while running faster and cheaper — reshaping the competitive landscape for AI-assisted development.GPT-4o — OpenAI's Multimodal Leap and What It Means for Developershttps://www.osmondvanhemert.nl/posts/240509-openai-gpt4o-multimodal/Thu, 09 May 2024 00:00:00 +0000https://www.osmondvanhemert.nl/posts/240509-openai-gpt4o-multimodal/OpenAI’s Spring Update reveals GPT-4o, a natively multimodal model that processes text, audio, and vision in a single architecture. The developer implications are significant.Meta Releases Llama 3 — Open Source AI Just Got Serioushttps://www.osmondvanhemert.nl/posts/240418-meta-llama-3-release/Thu, 18 Apr 2024 00:00:00 +0000https://www.osmondvanhemert.nl/posts/240418-meta-llama-3-release/Meta’s Llama 3 arrives with 8B and 70B parameter models that rival closed-source competitors, reshaping the open-weight AI landscape.Claude 3 Arrives — Anthropic's New Family of Models Raises the Barhttps://www.osmondvanhemert.nl/posts/240307-anthropic-claude-3-benchmarks/Thu, 07 Mar 2024 00:00:00 +0000https://www.osmondvanhemert.nl/posts/240307-anthropic-claude-3-benchmarks/Anthropic launches Claude 3 in three tiers — Haiku, Sonnet, and Opus — with benchmark results that challenge GPT-4’s dominance.Gemini 1.5 Pro — A Million Tokens Changes the Gamehttps://www.osmondvanhemert.nl/posts/240215-gemini-1-5-million-token-context/Thu, 15 Feb 2024 00:00:00 +0000https://www.osmondvanhemert.nl/posts/240215-gemini-1-5-million-token-context/Google’s Gemini 1.5 Pro launches with a 1 million token context window, fundamentally reshaping what’s possible with large language models.Google Rebrands Bard to Gemini — The AI Naming Game Gets Realhttps://www.osmondvanhemert.nl/posts/240208-google-gemini-rebrand-ai-platform/Thu, 08 Feb 2024 00:00:00 +0000https://www.osmondvanhemert.nl/posts/240208-google-gemini-rebrand-ai-platform/Google retires the Bard brand and goes all-in on Gemini. Behind the marketing refresh is a real technical shift with implications for developers building on Google’s AI stack.The GPT Store Is Live — What It Means for AI Developmenthttps://www.osmondvanhemert.nl/posts/240118-gpt-store-launch-ai-development/Thu, 18 Jan 2024 00:00:00 +0000https://www.osmondvanhemert.nl/posts/240118-gpt-store-launch-ai-development/OpenAI launches the GPT Store, creating a marketplace for custom GPTs. Here’s what it means for developers and why the platform play matters more than the individual bots.Google Gemini Arrives — Multimodal AI Gets Realhttps://www.osmondvanhemert.nl/posts/231207-google-gemini-multimodal-ai/Thu, 07 Dec 2023 00:00:00 +0000https://www.osmondvanhemert.nl/posts/231207-google-gemini-multimodal-ai/Google launches Gemini, its most capable AI model yet, bringing native multimodal reasoning to the forefront of the AI race.OpenAI DevDay — GPT-4 Turbo and the Platform Playhttps://www.osmondvanhemert.nl/posts/231109-openai-devday-gpt4-turbo/Thu, 09 Nov 2023 00:00:00 +0000https://www.osmondvanhemert.nl/posts/231109-openai-devday-gpt4-turbo/OpenAI’s DevDay unveils GPT-4 Turbo, custom GPTs, and the Assistants API — signaling a major shift from model provider to developer platform.Bletchley Park AI Safety Summit — Governments Finally Enter the Chathttps://www.osmondvanhemert.nl/posts/231102-bletchley-park-ai-safety-summit/Thu, 02 Nov 2023 00:00:00 +0000https://www.osmondvanhemert.nl/posts/231102-bletchley-park-ai-safety-summit/The UK’s AI Safety Summit at Bletchley Park brings 28 nations together to discuss AI risks, marking a watershed moment for international AI governance.Code Llama — Meta's Open Source Bet on AI-Assisted Codinghttps://www.osmondvanhemert.nl/posts/230824-code-llama-open-source-code-generation/Thu, 24 Aug 2023 00:00:00 +0000https://www.osmondvanhemert.nl/posts/230824-code-llama-open-source-code-generation/Meta releases Code Llama, a family of open-source code generation models, and it might just change the dynamics of AI-assisted development.Meta Releases Llama 2 — Open Source AI Gets a Massive Boosthttps://www.osmondvanhemert.nl/posts/230720-meta-llama2-open-source-llm/Thu, 20 Jul 2023 00:00:00 +0000https://www.osmondvanhemert.nl/posts/230720-meta-llama2-open-source-llm/Meta’s release of Llama 2 as a commercially-licensed open model changes the game for developers building with large language models.GPT-4 Lands — And It Raises the Bar Significantlyhttps://www.osmondvanhemert.nl/posts/230316-gpt4-lands-and-raises-the-bar/Thu, 16 Mar 2023 00:00:00 +0000https://www.osmondvanhemert.nl/posts/230316-gpt4-lands-and-raises-the-bar/OpenAI releases GPT-4 with multimodal capabilities and dramatically improved reasoning — here’s what it means for developers.The AI Search Wars Begin — Bing Chat, Google Bard, and the Future of Finding Thingshttps://www.osmondvanhemert.nl/posts/230209-ai-search-wars-bing-bard/Thu, 09 Feb 2023 00:00:00 +0000https://www.osmondvanhemert.nl/posts/230209-ai-search-wars-bing-bard/Microsoft and Google are racing to integrate large language models into search, and the implications go far beyond just finding web pages.ChatGPT's First Month — Why This AI Moment Feels Differenthttps://www.osmondvanhemert.nl/posts/221229-chatgpt-explosive-first-month/Thu, 29 Dec 2022 00:00:00 +0000https://www.osmondvanhemert.nl/posts/221229-chatgpt-explosive-first-month/One month after launch, ChatGPT has crossed a million users and sparked conversations about AI that reach far beyond the usual tech circles. Here’s why this one matters.GPT-3 API Access — First Impressions from the Betahttps://www.osmondvanhemert.nl/posts/200723-gpt3-api-beta-first-impressions/Thu, 23 Jul 2020 00:00:00 +0000https://www.osmondvanhemert.nl/posts/200723-gpt3-api-beta-first-impressions/OpenAI is granting beta access to the GPT-3 API. After a week of experimentation, here’s what’s genuinely impressive and what’s overhyped.Reformer — Can We Make Transformers Practical for the Rest of Us?https://www.osmondvanhemert.nl/posts/200123-reformer-efficient-transformers/Thu, 23 Jan 2020 00:00:00 +0000https://www.osmondvanhemert.nl/posts/200123-reformer-efficient-transformers/Google’s new Reformer model tackles the massive memory and compute costs of Transformers. For engineers building AI-powered features, this matters more than another benchmark score.