Problem

gpt-5-mini (configured in .local/config/testing.toml and likely in production configs) is missing from default_pricing() in crates/zeph-core/src/cost.rs. This causes a WARN on every LLM call:

WARN zeph_core::cost: model not found in pricing table; cost recorded as zero model="gpt-5-mini"

Cost is silently recorded as $0.00, making session cost tracking inaccurate.

Fix

Add gpt-5-mini to default_pricing() with the correct per-1K token rates. Also consider adding a gpt-5 entry for the full model.

Observed

Every test session using testing.toml since v0.14.x. High noise-to-signal ratio in logs.

Severity

Low — no functional impact, cost tracking only.