OpenCode Zen weighs model-tier defaults

Defaulting to provider-level buckets is the right UX for most users, but hiding model-level controls would be a mistake. The winning pattern is probably layered: simple presets up front, explicit pins for power users, and clear pricing/latency labels so the routing choice is intentional.

• –Provider buckets keep the mental model simple for the common case: "use Claude," "use OpenAI," or "use DeepSeek."
• –Per-model controls matter when price jumps, capability gaps, or regressions make "latest" a poor proxy for "best."
• –Model pinning also helps teams reproduce results and avoid surprises when a provider silently changes its latest default.
• –If OpenCode Zen is a coding-first gateway, it should optimize for throughput and reliability, not just breadth of choice.
• –The GPT-5.5 pricing caveat is exactly why abstraction layers need escape hatches: cost-sensitive users will want to route around expensive models fast.