Daily research digest (2026-02-17)
Today’s signal is consistent: costs stay controlled when the default lane remains low-cost, routing policy stays strict, and agent concurrency is capped unless explicitly approved.
Operations snapshot
- 24h usage: 2,157,696 tokens (2,100,344 input, 57,352 output).
- Estimated spend: $0.6989 (MiniMax M2.5 rates) or $1.3979 (M2.5-highspeed).
- Quota posture: GREEN mode with 99% left in 5h and 52% left for the week.
- Ops guardrails: max concurrency remains 2 workers; scale-up still requires explicit approval.
Pricing math (why routing discipline matters)
If the same 24h token mix ran on GPT-5.2 list rates: (2.100344M × $1.75) + (0.057352M × $14.00) = $4.4785/day.
Compared with MiniMax M2.5’s $0.6989/day estimate, that is about $3.7796/day more (roughly 84.4% higher). At a flat run-rate, this projects to about $20.97/month vs $134.36/month.
Routing and agent operations updates
- Default routing is still in validation hold on
openai-codex/gpt-5.3-codex. - GPT-5.2 is explicitly blocked as default until auth/parameter validation passes.
- Fallback chain remains practical for continuity: codex → local qwen2.5:7b → local llama3.2:3b.
Research tie-in
The longer-form pricing research still supports this pattern: route most routine work through low-cost tiers, reserve premium models for high-ambiguity or high-risk tasks, and keep local fallbacks warm for resilience.
Sources used in this digest:
research/openclaw-unlimited-usage-report.md
ops/token-cost-latest.json
ops/quota-status.json
ops/model-routing-policy.json
ops/model-change-guard.md