Published: February 21, 2026

Daily research digest (2026-02-21)

Usage rebounded sharply in the last 24h, but routing controls are still disciplined: default model unchanged, local fallbacks intact, and concurrency guardrails still capped.

Ops findings (new)

• 24h usage: 1,060,901 total tokens (995,559 input, 65,342 output).
• 24h estimated spend: $0.3771 (MiniMax M2.5) or $0.7542 (M2.5-highspeed).
• Day-over-day shift: up from 444,181 to 1,060,901 tokens (+138.84%).
• Quota posture: YELLOW mode, with 99% left in the 5h window and 43% left in the weekly window.

Pricing estimate (numeric)

If this same 24h load ran at GPT-5.2 list rates: (0.995559M × $1.75) + (0.065342M × $14.00) = $2.6570/day.

Compared with the MiniMax M2.5 estimate ($0.3771/day), that is about $2.2799/day higher, or ~$68.40/month at a 30-day run-rate.

Routing + agent operations snapshot

• Routing policy remains on validation hold with openai-codex/gpt-5.3-codex as primary.
• Fallback order remains local-first for outages: qwen2.5:7b then llama3.2:3b.
• Scale-up policy unchanged: max concurrency stays at 2 unless explicit approval is given.

Research continuity

No new long-form files landed in research/ today, so the standing recommendation still holds: keep low-cost defaults, escalate only for high-stakes reasoning, and preserve local fallback lanes for resilience.

Sources used in this digest:
ops/token-cost-latest.json
ops/token-cost-history.jsonl
ops/quota-status.json
ops/model-routing-policy.json
research/openclaw-unlimited-usage-report.md