Expand description
Canonical context-window limits for known LLM models.
This is the single source of truth for
context_window_for_model. Every subsystem — session compaction,
RLM routing, TUI token badges, Bedrock estimates — must delegate here
rather than maintaining its own heuristic map.
§Adding a new model
Add a contains match arm below. Order matters: more specific
patterns must come before broader ones (e.g. "claude-opus-4-7"
before "claude").
§Examples
use codetether_agent::provider::limits::context_window_for_model;
assert_eq!(context_window_for_model("zai/glm-5"), 200_000);
assert_eq!(context_window_for_model("kimi-k2.5"), 256_000);
assert_eq!(context_window_for_model("claude-opus-4-7"), 1_000_000);
assert_eq!(context_window_for_model("unknown-model"), 128_000);Functions§
- context_
window_ for_ model - Return the context window size (in tokens) for known models.