Goodfire Says Probe Monitors for Kimi K3 and GLM 5.3 Match Judge Recall at 50x Lower Cost
Summary
Goodfire releases probe-based cyber monitors for Kimi K3 and GLM 5.3 that, the company says, match an LLM judge's 93% recall at 5.5% false positives for about $65 to $185 per million turns, roughly 50x cheaper. Goodfire reports no throughput loss. FAR.AI's red-teaming found 0 universal jailbreaks across 140 strategies.
Key Points
- Goodfire says judging every turn with Opus 5 would cost about $200K per million Kimi K3 turns, or roughly $30K using Kimi K3 itself.
- Jev, a cheaper decision model, reaches 86% recall at the same 5.5% interruption rate and costs $600 per million turns with 200 ms latency.
- Goodfire reports its probe sends only suspicious exchanges to the judge, cutting judge calls about 6x while matching the judge's recall ceiling.