Flagship tier

Most queries don't need the top of the line. Some do. The trick is knowing which — and BEAM makes the split explicit rather than hiding it in a pricing page.

Standard vs Flagship

Every BEAM account runs on standard models: Claude Sonnet, GPT's current mainline, Gemini Flash. These are fast, inexpensive, and right for the overwhelming majority of what a workspace does — parsing your phrasing, answering grounded questions, running a council on a decision.

The flagship tier swaps each provider's standard model for its frontier one — Claude Opus, OpenAI's top reasoning model, Gemini Pro. Frontier models are several times more expensive per token and noticeably slower. What you buy is depth: better performance on long chains of reasoning, technical primitives, and arguments where a subtle flaw costs you weeks.

During the private beta, flagship access is granted per-account. If Settings shows an AI Models tab, your account has it; if not, it hasn't been enabled for you yet — everything else in this piece still describes how BEAM behaves once it is.

When to flip it on

The pattern that pays for flagship: decisions you'll act on for weeks. Architecture choices. Strategy documents. Anything involving security, cryptography, statistics, or legal structure — domains where a confident-sounding standard answer can be subtly wrong in ways you can't check. If you find yourself planning to trust the answer without verifying it, that's the signal to elevate.

The pattern that doesn't: routine throughput. Daily briefs, task parsing, quick lookups, first drafts. The standard tier is genuinely good at these, and flagship's extra depth mostly buys you latency.

Per-call vs persistent

Elevation comes in two grains. --flagship on a single command (ask --flagship, council --flagship) elevates just that call. model flagship on makes it persistent until you turn it off — the status bar shows an amber ⚡ so you always know you're spending at the higher rate. Every response's usage line shows the model and cost either way, so the spend is never invisible.

Looking ahead

The manual toggle is the honest interim: you know which questions matter, so you decide. The longer arc is learned routing — BEAM recognizing the questions that deserve frontier depth and offering the elevation itself. Until then, the rule of thumb above is the router.

See also: Reference: Settings · model · Concepts: Council vs Debate vs Wild