Understanding Logprobs: What They Are and How to Use Them
Learn what OpenAI's logprobs are and how can you use them for your LLM applications
← Back to blog
Document Data Extraction in 2026: LLMs vs OCRs
A choice dependent on specific needs, document types and business requirements.
GPT-5 Benchmarks
See how GPT-5 performs across benchmarks; with a big focus on health
Google's AP2: A new protocol for AI agent payments
How verifiable mandates are creating a secure foundation for AI-driven commerce.
A Guide to LLM Observability
Think your APM tool has your AI covered? Think again. LLMs need their own observability playbook.
OpenAI's Agent Builder Explained
A breakdown of OpenAI’s new Agent Builder and what it signals for the future of building and deploying AI agents.
Zero-Shot vs Few-Shot prompting: A Guide with Examples
Exploring zero-shot & few-shot prompting: usage, application methods, and limits.
Chain of Thought Prompting (CoT): Everything you need to know
We break down when Chain-of-Thought adds value, when it doesn’t, and how to use it in today’s LLMs.
Build AI Products Faster: Top Development Platforms Compared
Compare top AI platforms for fast, reliable development in 2025.
Understanding your agent’s behavior in production
You can’t improve what you can’t see, so start tracking every decision your agent makes.
How can agentic capabilities be deployed in production today?
A practical guide to deploying agentic capabilities: what works, what doesn’t, and how to keep it reliable in prod.
Partnering with Composio to Help You Build Better AI Agents
Building AI agents is 10x easier with 10,000+ tools and built-in LLM tooling support
OpenAI o3 vs gpt-oss 120b
Just another eval confirming 90% discount with highest performance from GPT-OSS 120b.
How to craft effective prompts
A curated list of best practices, techniques and practical advice on how to get better at prompt engineering.
Subliminal Learning in LLMs
LLMs carry hidden traits in their data and we have no idea how.
Introducing Vellum Agent Builder
Go from idea to AI workflow in seconds and continue to build in the UI or your IDE.
Introducing Custom Docker Images & Custom Nodes
Complete control over the business logic and runtime of your AI workflows in Vellum.
Big Ideas from the AI Engineer World’s Fair
What’s shaping AI products, agents, and infrastructure in 2025.
10 Humanloop Alternatives in 2025
A side-by-side look at Humanloop and 10 other LLM platforms.
Evaluation: Claude 4 Sonnet vs OpenAI o4-mini vs Gemini 2.5 Pro
Analyzing the difference in performance, cost and speed between the world's best reasoning models.
How to connect a Vellum AI Workflow with your Lovable app
Build a functional chatbot using Vellum AI Workflows and Lovable with just a few prompts.
How to evaluate an LLM evaluation framework
A quick guide to picking the right framework for testing your AI workflows.
Evaluating models on adaptive reasoning, SAT questions & real-world classification tasks
Evaluating SOTA models if they can really reason
Four Reasons Enterprise AI Projects Get Stuck
A wake up call to not underestimate the unique challenges of working with LLMs.
MCP: The Hype vs. Reality
LLMs are stepping outside the sandbox. Should you let them?
How Drata built an enterprise-grade AI solution with Vellum
See how Drata leveraged Vellum to build enterprise-grade AI workflows that enhance GRC automation.
Native integration with IBM’s Granite models
Support for IBM granite models in Vellum.
GPT-4.5 vs Claude 3.7 Sonnet
Comparing GPT-4.5 and Claude 3.7 Sonnet on cost, speed, SAT math equations, and adaptive reasoning skills.
GPT 4.5 is here: Better, but not the best
Feels more natural, hallucinates less, can be persuaded—and it’s not a game-changer.
Claude 3.7 Sonnet vs OpenAI o1 vs DeepSeek R1
Learn how the latest Anthropic's model compares to similar top-tier reasoning models on the market.
How RelyHealth Deploys Healthcare AI Solutions 100x Faster
Learn how Vellum enables Rely Health to rapidly build, test, and deploy AI-powered patient care solutions.
How Revamp Reliably Runs 15M+ LLM Executions in Production
Learn how to optimize prompt versioning, debug efficiently, and make real-time updates to boost AI performance.
Claude 3.7 Sonnet: Can It Actually Reason?
Evaluating the 'thinking' of Claude 3.7 Sonnet and other reasoning models to understand how they really reason.
Analysis: OpenAI o1 vs DeepSeek R1
Explore how O1 and R1 perform on well-known reasoning puzzles—now tested in new contexts.
Breaking down the DeepSeek-R1 training process—no PhD required
Learn how DeepSeek achieved OpenAI o1-level reasoning with pure RL and solved issues through multi-stage training.
What to do when an LLM request fails
Rate limiting and downtime are common issues with LLMs — here’s how to manage it in production.
Llama 3.3 70b vs GPT-4o
Learn how the latest model from Meta, Llama 3.3 70b compares to GPT-4o on three tasks
Native support for SambaNova inference in Vellum
Now you can run Llama 3.1 405b, with 200 t/s via SambaNova on Vellum!
AI Development Survey: Help us build the ultimate AI changelog
Share your AI process in our 4-minute anonymous survey. Get early insights and a chance to win a MacBook M4 Pro.
Announcing Native Support for Cerebras Inference in Vellum
Starting today, you can unlock 2,100 t/s with Llama 3.1 70B in Vellum for real-time AI apps.
How Glowing Personalized Hospitality Experiences with AI
Discover how Glowing leverages Vellum's Workflows to create innovative AI solutions for the hospitality industry.
OpenAI o1: Prompting Tips, Limitations, and Capabilities
Learn how to prompt OpenAI o1 models, understand their limits and the opportunities ahead.
LLM Benchmarks: Overview, Limits and Model Comparison
Understand the latest benchmarks, their limitations, and how models compare.
How Woflow Decoupled AI Updates for 50% Faster Delivery — Without the Infra Stress
Learn how Woflow sped up AI development by 50% — making it easier to handle errors, improve models and ship updates.
How this EdTech Company Made AI Development 10x Faster with Vellum
Explore how a leading EdTech company saves 50 eng hours per month and empowers everyone on the team to contribute.
The 6 Stages for Successful AI Implementation
Learn critical strategies to build and launch AI systems quickly and reliably.
How Vellum Helped Odyseek Build Smarter AI Faster
Learn how Odyseek used Vellum to simplify AI development and improve team collaboration.
Llama 3.1 405b vs Leading Closed-Source Models
Discover How Llama 3.1 405b Stacks Up Against GPT-4o, Gemini 1.5 Pro, and Claude 3.5 Sonnet on Three Tasks
Evaluation: Llama 3.1 70B vs. Comparable Closed-Source Models
Explore Llama 3.1 70b's upgrades and see how it stacks up against same-tier closed-source models.
Claude 3.5 Sonnet vs GPT-4o
Learn how Claude 3.5 Sonnet compares to GPT4o on data extraction, classification and verbal reasoning tasks.
Llama 3 70B vs GPT-4: Comparison Analysis
Find out how Llama 3 70B stacks up against GPT-4 in terms of cost, speed, and performance on specific tasks.
Rentgrata's Test Driven Journey to a Production-Ready Chatbot
Learn how Rentgrata used Vellum to evaluate their chatbot, and cut development time in half.
LlamaIndex vs LangChain Comparison
Discover what are the main differences between LangChain and LlamaIndex, and when to use them.
RAG vs Fine-Tuning: How to Choose the Right Technique?
Learn how RAG compares to fine-tuning and the impact of both model techniques on LLM performance.
Tutorial: Setting Up OpenAI Function Calling with Chat Models
Learn how to use OpenAI function calling in your AI apps to enable reliable, structured outputs.
How Autobound Achieved a 20x Faster End-to-End LLM Iteration Cycle
Iterating on prompts using OpenAI's playground & Azure AI studio was challenging, until Autobound discovered Vellum.
Redfin's Test Driven Development Approach to Building an AI Virtual Assistant
Discover how Redfin used Vellum to develop and evaluate a production-ready AI assistant, now live in 14 markets.
How to Count Tokens Before you Send an OpenAI API Request
Learn how to use Tiktoken and Vellum to programmatically count tokens before running OpenAI API requests.
Getting Started with Prompt Chaining
Learn how to improve LLM outputs, and make your setup more reliable using prompt chaining.
How to Evaluate Your RAG System?
Learn how to use retrieval and content generation metrics to consistently evaluate and improve your RAG system.
How can I get GPT-3.5 Turbo to follow instructions like GPT-4?
Learn prompt engineering tips on how to make GPT-3.5 perform as good as GPT-4.
How Lavender cut latency by half for 90K monthly requests in production
Learn how Lavender develops and manages more than 20 LLM features in production.
Prompt Engineering Guide for Claude Models
Learn how to prompt Claude with these 11 prompt engineering tips.
How Codingscape improved time-to-market for their AI apps
Learn how Vellum helped Codingscape to ship AI apps quicker and win more projects.
How can I use LLMs to classify user intents for my chatbot?
Learn how to build and evaluate intent handler logic in your chatbot workflow
3 Strategies to Reduce LLM Hallucinations
Methods and techniques to reduce hallucinations and maintain more reliable LLMs in production.
Four LLM hallucinations and ways to fix them
What is LLM hallucination & the four most common hallucination types and the causes for them
Classifying Customer Tickets using Gemini Pro
Comparing the performance of Gemini Pro with zero and few shot prompting when classifying customer support tickets
Best Model for Text Classification: Gemini Pro, GPT-4 or Claude2?
Comparing GPT3.5 Turbo, GPT-4 Turbo, Claude, and Gemini Pro on classifying customer support tickets.
Tree of Thought Prompting: What It Is and How to Use It
Learn how to use Tree of Thought prompting to improve LLM results
User Confidence in OpenAI vs. Alternative models/providers
Discover how recent OpenAI developments have influenced user confidence and interest in OpenAI alternatives
First impressions with the Assistants API
Assistants API: Easy assistant setup with memory management - but what's under the hood?
The ABC’s of Multimodal AI: Models, tasks and use-cases
How to use Multimodal AI models to build apps that solve new tasks and offer unique experiences for end users.
Automatic data labeling with LLMs
LLMs can label data at the same or better quality compared to human annotators, but ~20x faster and ~7x cheaper.
How Narya's team uses Vellum for auto data labeling & deployments
Learn how Vellum helped Narya.AI save time and make AI easy for everyone on their team.