---
type: "post"
id: "baseten-launches-grid-a-gateway-and-router-for-956bed"
slug: "baseten-launches-grid-a-gateway-and-router-for-956bed"
title: "Baseten launches Grid, a gateway and router for enterprise coding agents"
description: "Baseten opened public preview of Grid, an enterprise control layer combining a model gateway, task-level routing and central guardrails for coding agents."
retrieval_nugget: "Baseten opened public preview of Grid, an enterprise control layer combining a model gateway, task-level routing and central guardrails for coding agents. It gives platform teams one place to govern model access and usage across familiar harnesses; Baseten says early benchmarks reduced cost per coding task by nearly 70% with a small quality tradeoff."
published_at: "2026-11-05"
updated_at: "2026-11-07"
record_date: "2026-11-05"
date_kind: "published_at"
topics: ["coding-agents","claude","claude-code"]
entities: ["Baseten","Baseten Grid","Baseten Router","Claude Code","Codex","OpenCode","Pi","Opendoor"]
related_patterns: ["agent-economics-moves-to-completed-work","harness-architecture-outlives-model-choice"]
novelty_label: "same_event"
prior_refs: ["https://newruntime.com/posts/baseten-introduces-grid-sandboxes-and-serverless-loops-5-aa6489/"]
source_url: "https://www.baseten.co/blog/introducing-baseten-grid"
source_urls: ["https://www.baseten.co/blog/introducing-baseten-grid"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:baseten-launches-grid-a-gateway-and-router-for-956bed"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/baseten-launches-grid-a-gateway-and-router-for-956bed/","markdown":"https://newruntime.com/posts/baseten-launches-grid-a-gateway-and-router-for-956bed.md","json":"https://newruntime.com/posts/baseten-launches-grid-a-gateway-and-router-for-956bed.json"}
markdown_url: "https://newruntime.com/posts/baseten-launches-grid-a-gateway-and-router-for-956bed.md"
---

# Baseten launches Grid, a gateway and router for enterprise coding agents

## Retrieval answer

Baseten opened public preview of Grid, an enterprise control layer combining a model gateway, task-level routing and central guardrails for coding agents. It gives platform teams one place to govern model access and usage across familiar harnesses; Baseten says early benchmarks reduced cost per coding task by nearly 70% with a small quality tradeoff.

## Summary
Baseten opened public preview of Grid, an enterprise control layer combining a model gateway, task-level routing and central guardrails for coding agents. It gives platform teams one place to govern model access and usage across familiar harnesses; Baseten says early benchmarks reduced cost per coding task by nearly 70% with a small quality tradeoff.

## What happened
Baseten announced Grid on November 4, 2026, describing it as a way for enterprises to route requests, monitor usage and govern access across AI coding workflows. Grid is in public preview. The source presents three capabilities: Baseten Router for intelligent routing, a Gateway for model access and usage controls, and Guardrails for centrally configured safety policies and risk visibility.

Baseten Router selects among models approved by an organization, weighing expected task performance and cost. The company says its approach considers more than token prices: it includes the cost of retries, failed tool calls and incomplete answers. Teams can set requirements around cost, intelligence or a balance of the two. In initial benchmarks, Baseten reports nearly 70% cost savings with a 0.9-percentage-point quality tradeoff versus Opus 5.5. The company cautions that actual savings depend on workload shape and the desired balance between cost and quality. Organizations may opt in to sharing traces, which Baseten says can be used to train custom routers for their workloads.

The Gateway provides a single place to approve models, manage access, set budgets and view usage across teams, applications and agents. The announcement lists models from OpenAI, Anthropic, Gemini Enterprise Agent Platform, Amazon Bedrock and Baseten. Developers can connect Claude Code, Codex, OpenCode and Pi with a single setup command; administrators may automate configuration through mobile device management. In-house applications and agents that use the Anthropic or OpenAI API can connect through a base URL and model ID. Platform teams can set default models for coding and smaller models for background tasks, while developers can change models later.

Grid’s Guardrails apply configurable safety policies at the gateway layer so requests and responses are subject to a common baseline. Baseten frames this as a response to risks from agents reading internal documents, calling tools and acting on production systems. The announcement names Opendoor as an early user. Its Head of Engineering says Grid provides visibility into internal AI spending and helps cut spend automatically. The supplied material does not state Grid’s price or preview eligibility criteria.

## What the archive adds
The archive covered Baseten’s earlier product milestone in [Baseten Introduces Grid, Sandboxes, and Serverless Loops](https://newruntime.com/posts/baseten-introduces-grid-sandboxes-and-serverless-loops-5-aa6489/). This announcement is a new event in that product thread: it specifies Grid’s gateway, routing and guardrail functions, reports an initial cost benchmark and names Opendoor as an early user.

## Trend signal
This supports [agent economics moving to completed work](https://newruntime.com/patterns/agent-economics-moves-to-completed-work/): Baseten says its router accounts for retries, failed tool calls and incomplete answers, and reports cost per coding task rather than only token prices. The evidence is an early vendor benchmark, so it illustrates the direction of measurement but does not establish that the savings generalize across workloads or providers.

## Why now
Baseten’s announcement ties the launch to rising enterprise use of coding agents and the resulting cost and governance burden. It says developers currently have to choose models themselves while platform teams manage multiple providers, harnesses, budgets and policies. Grid is positioned as a way to centralize those choices as model options expand.

## Connections to other events
An earlier Telegram post described competition among agent gateways over distribution, including one-command setup and model availability. Grid’s announcement adds a specific enterprise emphasis: routing across approved models, organization-level budgets and centralized guardrails, alongside compatibility with existing coding harnesses.

## Why it matters
For teams adopting coding agents, model choice is becoming an operational control, not just an individual developer preference. A gateway that preserves existing harnesses while centralizing credentials, budgets and approved models could make it easier to offer agents broadly without losing cost visibility or policy consistency. Task-level routing also reflects a practical question: whether a cheaper model can complete a particular job well enough, including the retries and tool failures that a token-price comparison misses.

The potential benefit is meaningful, but the evidence is not yet enough to treat the reported savings as a planning assumption. Teams would need to compare routing against their own workload, quality thresholds and failure costs, and verify that governance controls fit their security requirements. The announcement describes the product’s intended controls and an early customer account, not independent validation of either the benchmark or the guardrails. Earlier coverage: [Baseten Introduces Grid, Sandboxes, and Serverless Loops](https://newruntime.com/posts/baseten-introduces-grid-sandboxes-and-serverless-loops-5-aa6489/).

## Limits and uncertainty
The cost and quality figures come from Baseten’s initial benchmarks; the supplied announcement does not provide benchmark task counts, workload composition, evaluation methodology or independent results. The Opendoor account is a customer quotation rather than a quantified case study. The material does not provide pricing, preview eligibility, service limits, deployment details or a fuller technical description of the guardrails. It also does not establish that routing will preserve quality for any specific team’s tasks, or how the listed providers and models compare under a common evaluation.

## Skill to build now
Before enabling automatic routing, assemble a small representative set of coding-agent tasks and define acceptance criteria for each: successful completion, tool-use correctness and acceptable review effort. Record model, retries, failures, elapsed time and total cost per completed task. That gives a team a baseline to compare against a router’s claims and to decide which work can safely use a cheaper model.

## Reader takeaway
Grid’s public preview packages model access, routing and policy controls around existing coding-agent harnesses. Its central proposition is to optimize cost per task while keeping quality within a team’s requirements. The nearly 70% savings figure is an early Baseten result, not a universal expectation; the useful next step for a buyer is to test routing against real work and explicit acceptance criteria.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Coding Agents](https://newruntime.com/topics/coding-agents.md): Explore the coding-agents topic hub.
- topic hub: [Claude](https://newruntime.com/topics/claude.md): Explore the claude topic hub.
- topic hub: [Claude Code](https://newruntime.com/topics/claude-code.md): Explore the claude-code topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Sources](https://newruntime.com/sources/): See the original sources this archive cites.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)