Weighted criteria
9.3/10Scoring breakdown
AI Models weights turn category-specific evidence into the score.
Reasoning reliability35% weight9.7/10
Strongest score because it keeps constraints intact across code, research, and planning work.
Weighted impact: 3.40/10Tool and workflow fit25% weight9.4/10
Works best in review gates and agent plans where careful tool use matters more than speed.
Weighted impact: 2.35/10Speed and cost control20% weight8.4/10
Premium latency and price need routing rules, so it loses points as a blanket default.
Weighted impact: 1.68/10Context and privacy fit20% weight9.2/10
Long briefs and hosted controls are useful when teams already have data-handling rules.
Weighted impact: 1.84/10Route hard code review, architecture, agent planning, and research synthesis here first.
Do not use it as the blanket default for extraction, summaries, simple support replies, or bulk drafts.
Wait if your product cannot tolerate slower first-token response or you have no router to contain cost.
Measured fit
- Context fit
- Long technical briefs
- Latency class
- Deliberate
- Cost shape
- Premium per solved hard task
- Privacy posture
- Hosted controls
- Reliability
- Excellent
- Speed
- Moderate
- Cost discipline
- Needs routing
Evidence and caveats
- Handled multi-file review prompts with fewer invented fixes than faster utility models.
- Maintained constraints across long plans instead of optimizing the last instruction only.
- Became cheaper per solved hard task when used behind an escalation rule.
Too slow for high-volume repliesNeeds routing rules to keep spend controlled