From the source
Baseten opened a public preview of Grid, a routing, gateway and guardrails layer that picks which approved model handles each enterprise coding task, with early access through a waitlist.
- Baseten reported that its Router achieved nearly 70% savings in cost per task in initial benchmarks, with a 0.9-percentage-point quality tradeoff against Opus 5.5.
- The router weighs retries, failed tool calls and incomplete answers rather than ranking models by token price alone, and routes can emphasize cost, intelligence or a balance of both.
- Organizations can opt in to sharing traces that train custom routers for Pareto efficiency on their workloads; Baseten said exact savings vary with workload shape and the desired cost-quality balance, so the benchmark indicates potential rather than a universal reduction.
- Grid Gateway lets platform teams approve models, grant team-level access, set budgets and use company-managed credentials, with a dashboard breaking down spend by model alongside token volume, cache hit rate and cost saved.
- At launch Grid supports models from
OpenAI,
Anthropic, Gemini Enterprise Agent Platform,
Amazon Bedrock and Baseten, and developers can connect Claude Code, Codex, OpenCode and Pi with one setup command.
- Configurable safety policies sit at the gateway, and Baseten plans support for delegating longer-running work to cloud-hosted agents, coordinating multiple agents, routing tuned to customer data and support for customers' own guardrails.