LangChain vs LiteLLM

Versos AI Launches NVIDIA NeMo Agents for Video Training Data Curation

Versos AI, a provider of video training data infrastructure for library owners and AI labs, announced a new capability built with NVIDIA NeMo. This allows AI teams to describe their dataset needs in natural language, which the platform then uses to search, evaluate, and assemble rights-cleared video footage into structured training datasets. The workflow accelerates model development by reducing manual search and review time. It also enhances discoverability of premium video libraries for AI partnerships. The technology leverages NVIDIA CUDA for accelerated inference, NVIDIA Nemotron Ultra for agent reasoning, and LangChain for agent orchestration. Versos AI will demonstrate this at IBC2026 in Amsterdam.

• Versos AI launched a new capability built with NVIDIA NeMo.
• The system interprets natural language requests to find and assemble video training data.
• Built on NVIDIA CUDA Toolkit, NVIDIA Nemotron Ultra, and LangChain.

LangGraph Outperforms CrewAI and AutoGen in Data Engineering Benchmark

A developer benchmark on 107 real data engineering tasks compares LangGraph, CrewAI, and AutoGen. LangGraph achieves the highest pass rate (97/107) with lower latency and token usage. CrewAI shows higher token consumption and latency, while AutoGen struggles with stateful multi-step operations, leading to frequent failures. The article provides code examples and operational metrics, concluding that LangGraph's explicit graph-based control flow is more reliable and cost-efficient for agentic ETL pipelines.

• LangGraph passed 97/107 tasks, CrewAI 80/107, AutoGen 58/107.
• LangGraph median token usage per task: 2350; CrewAI: 4120; AutoGen: 3160.
• LangGraph mean latency: 13.8s; CrewAI: 21.6s; AutoGen: 29.2s.

RAG's Forgotten Foundation: Study Information Retrieval

The article argues that Retrieval-Augmented Generation (RAG) is essentially a classic search engine with an LLM appended, and that modern RAG projects fail when teams rely solely on vector embeddings and expensive infrastructure. It recommends re-learning Information Retrieval (IR) fundamentals—lexical search (BM25), hybrid search, multi-stage retrieval pipelines (cheap retrievers + expensive re-rankers), and rigorous IR evaluation metrics (Precision@K, Recall, NDCG)—to reduce cost, improve robustness to embedding model drift, and scale to large data volumes. The author points readers to the textbook Introduction to Information Retrieval (Manning, Raghavan, Schütze) as a practical source of foundational techniques that can make production RAG systems more reliable and affordable.

• Author argues RAG is effectively a classical search engine with a large language model attached.
• The article recommends BM25 (lexical search) as a stable, disk-based retrieval method using inverted indexes.
• Industry practice promoted: Hybrid search (vector retrieval + BM25) to mitigate embedding model drift.

Auto Router: 45% Lower Cost on 25 SWE-bench Tasks

We solved 23 of 25 SWE-bench Verified tasks with LiteLLM's experimental capability router for $11.15, compared with $20.27 using Opus 5.

Introducing LiteLLM Fusion: 56% More Tasks Solved Than Fable 5

LiteLLM Auto Router Fusion ran three models on the same task and synthesized their work, solving 14 of 21 Terminal-Bench tasks against 9 for Claude Fable-5 alone. Total spend rose 36%, cost per solved task fell 12%, and turn latency went up 5x.

Configure LiteLLM as Codex Model Provider

A developer guide demonstrating how to route Codex to use LiteLLM as a custom model provider. The post explains exposing a LiteLLM API key as an OS environment variable, updating Codex's .codex/config.toml to set model_provider to 'litellm' and add provider-specific fields (base_url, env_key, wire_api, streaming options), setting optional custom HTTP headers, and noting that session models are fixed at session creation. The author also advises verifying usage via LiteLLM dashboard logs and links to LiteLLM and Codex documentation for reference.

• Author Julia Shevchenko published a how-to on dev.to on 2026-08-28 about configuring LiteLLM as a gateway for Codex.
• LiteLLM exposes an OpenAI-compatible interface and can act as a gateway for LLMs.
• Required steps include setting LITELLM_API_KEY as an OS environment variable and updating .codex/config.toml to set model_provider = "litellm" and provider-specific settings (base_url, env_key, wire_api, streaming options).