A context window measured in codebases, not pages
Bee Comb and Buzz serve a dedicated 1M-token long-context mode. Swarm runs 1M natively. Every paid tier starts at a 256K-token native window.
The context ladder, stated exactly
These are served figures, not aspirational maxima: the long-context mode is provisioned as its own serving path, sized and verified for full-window requests, rather than a config flag stretched past what the deployment can hold.
- →Bee Cell (free) — 128K-token context window.
- →Bee Brood, Comb, Buzz, and Hive — 256K-token native context.
- →Bee Comb and Buzz — dedicated 1M-token long-context serving mode.
- →Bee Swarm — 1M-token context natively.
What fits in one million tokens
A 1M-token window is roughly four times the 256K native window — enough headroom to hold a large service's source tree, a multi-year contract set, or a quarter's meeting transcripts in a single request and ask questions that span all of it.
That changes how you work: instead of engineering a pipeline to decide what the model is allowed to see, you load the corpus and ask.
Long context and retrieval are complements
Retrieval excels when the source set is far larger than any window or changes continuously; long context excels when reasoning must span everything at once — cross-file refactors, contradiction hunting across contracts, timeline reconstruction across transcripts.
Bee ships both on the same platform: retrieval with per-tenant indexes and cite-back grounding, and the 1M window when you need the whole corpus in working memory. You choose per request, not per vendor.
Same API, bigger window
Long-context requests use the same OpenAI-compatible Chat Completions endpoint — no separate product, no new client. Token accounting stays transparent: usage reports the real prompt and completion counts, and pooled-token plans meter input and output 1:1.
Frequently asked questions
- Which Bee tiers serve the 1M-token window?
- Comb and Buzz serve a dedicated 1M-token long-context mode, and Swarm serves 1M natively. The other paid tiers run 256K native context, and the free Cell tier runs 128K.
- Is the 1M window real or a marketing maximum?
- It's served capacity on a dedicated long-context path, verified with full-window requests — not a theoretical config value. Bee's policy is that published context figures match what the deployment actually serves.
- When should I use RAG instead of long context?
- Use retrieval when the corpus is much larger than 1M tokens or updates continuously, and long context when the question needs simultaneous reasoning over everything loaded. Bee supports both on the same endpoint.
- Does long context cost more?
- You pay for the tokens you actually send at the tier's published per-token rate — a bigger window raises the ceiling, not the price per token. Workspace plans meter long-context requests against the same pooled allowance.
- How do I send a long-context request?
- The same way as any Chat Completions call — the window is a property of the tier you select, so no special request format is needed.
Related
Start on the free tier
Bee Cell is free — no card. Scale to paid tiers, the API, or sovereign deployment when you are ready.