How Castform + Neon Beats Frontier Models on Price and Efficiency
- ID: 9c5366ef
- Original URL: https://neon.com/blog/how-castform-neon-beats-frontier-models-on-price-and-efficiency
- Author(s): Pranav Aurora
- Date: 2026-08-05
- Category: infra
- Source type: article
- Tags: agentic-retrieval, post-training, neon, cost-efficiency, rl
- Quality score: 4/5
- Fetched at: 2026-08-06T15:43:42+00:00
- Obsidian evidence: OpenClaw定时任务/ClawFeed24小时高价值一览/2026-08-06-ClawFeed24小时高价值一览.md
中文导读
Neon / Castform 把 agentic multi-hop 检索成本问题拆成可训练、可算账的系统问题：私有语料进入 Neon Postgres + Lakebase hybrid search，用合成问答任务和检索/引用/正确性奖励对 4B 开源模型做任务向 RL post-train。文章称检索准确率可接近 frontier 多轮搜索，而推理成本低约两个数量级；同时强调 stateful agent 训练需要廉价 branch 与隔离环境。
为什么值得关注
对私有知识库 agent 的成本、训练和数据库隔离有实际参考价值。
English Summary
The post describes using Neon Postgres/Lakebase hybrid search plus Castform task-oriented RL post-training on a 4B open model to approach frontier retrieval behavior at substantially lower inference cost, with cheap branching for stateful training.
原文摘要 / Source Excerpt
How Castform + Neon Beats Frontier Models on Price and Efficiency
作者: Pranav Aurora
发布时间: 2026-08-05T12:00:00
原文链接: https://neon.com/blog/how-castform-neon-beats-frontier-models-on-price-and-efficiency
How Castform + Neon Beats Frontier Models on Price and Efficiency
A 4B open-source model post-trained with Castform retrieved search results as accurately as GPT-5.6 Sol, while costing 100x less
Pranav Aurora, Ying Hang Seah, Angel Pan
Aug 05, 2026
Subscribe to our changelog
Receive only our latest updates. No spam, guaranteed.
“Most teams' best training data is just sitting in their databases. The problem is that turning raw data into something usable is hard, and letting agents read, search, and mutate data cheaply at scale requires advanced infra. Pointing Castform at Neon skips both.”
Ying Hang Seah, cofounder, Castform
A "good agent" needs to be strong in 2 areas:
- Context: can we provide the tools to find the right data?
- Model: can the model decide what to search for?
Neon (Lakebase Postgres) and their new Search extensions solve the first; Castform solves the second.
Evolution of agentic search#evolution-of-agentic-search
In ~2022, the industry was going all in on embedding search. Every database provider added one, and pgvector was Neon's most downloaded extension. To provide context to LLMs, engineers handcrafted RAG pipelines, which in essence, is some form of embedding similarity search.
In ~2025, agents started to gain more traction. Developers started creating multi-hop search workflows, decomposing big problems into smaller ones. Retrieval has shifted from the one-shot search systems to agentic retrieval. Instead of issuing a single query, models plan and search multiple times in a loop. Every loop iteration meant another call to the frontier model, increasing the overall cost and latency per user request.
Concretely, a typical multi-turn search request with gpt-5.6-sol takes >10s and costs ~$0.03 end-to-end, making it prohibitively slow and expensive.
Meanwhile, small open-weights models are 100x cheaper. But, out of the box, their capabilities lag behind closed api models. RL post-training helps bridge this gap. On specific tasks like search, post-trained open-source models can match & beat frontier models while costing orders of magnitude less per request.
That is why we built Castform: to enable developers to RL post-train models without having to deal with machine learning & gpu internals. The goal's to make post-training as approachable as prompt engineering.
How does Castform use Neon?#how-does-castform-use-neon
Castform's pipeline runs against Neon via Lakebase Search:
| Stage | Neon + Lakebase Search | | --- | --- | | Corpus storage | Raw documents live in Postgres on Neon | | Synthetic data generation | Castform training pipeline uses lakebase_text and lakebase_vector to write training tasks | | RL Training | Every rollout's search tool call uses Lakebase Search on Neon | | Production Inference | The final model uses the same search tool call during inference |
Your best training data already exists[](#your-best-training-data-already-exi
...[excerpt truncated, fetched body length=11958 chars]...