Fronset Benchmark

Tencent Hy3 at a glance

Good enough on 16/54 tasks at the 90% bar. Best value on 1 task. Best fit for: Topic Organization & Clustering. Doesn't qualify on any: Financial Analysis & Trading Decisions.

Provider
OpenRouter
Model name
tencent/hy3
Cost mode:
Qualifies on
16 / 54 tasks (at 90% bar)
Best value on
1 tasks

Cost vs quality across all tasks

0%25%50%75%100%78910Quality score (7–10)Cost-efficiency vs best value (1.0 = best value)Public Response Generation — quality 7.87, 9.1x the cost of the best-value good-enough optionTopic Sequence Optimization — quality 8.24, 3.2x the cost of the best-value good-enough optionModel Specific Prompt Adaptation — quality 8.50, 11.7x the cost of the best-value good-enough optionPromotional Message Generation — quality 7.78, 5.8x the cost of the best-value good-enough optionAtomic Fact Claim Extraction — quality 7.37, 10.4x the cost of the best-value good-enough optionBatch Text Translation — quality 8.37, 17.6x the cost of the best-value good-enough optionBench Hypothesis Post Generation — quality 8.30, 15.8x the cost of the best-value good-enough optionContent To Image Prompt Generation — quality 8.17, 11.4x the cost of the best-value good-enough optionFactual Claim Refinement — quality 8.27, 9.6x the cost of the best-value good-enough optionStructured Content Summarization — quality 8.69, 8.3x the cost of the best-value good-enough optionLanguage Identification — quality 9.95, 11.1x the cost of the best-value good-enough optionEngagement Opportunity Triage — quality 8.57, 10.7x the cost of the best-value good-enough optionTopic Section Assignment — quality 7.95, this model IS the best-value good-enough option★Social Post Portfolio Selection — quality 7.98, 1.3x the cost of the best-value good-enough optionTopic Cluster Labeling — quality 8.13, 7.3x the cost of the best-value good-enough optionVisual Theme Configuration Generation — quality 8.11, 3.2x the cost of the best-value good-enough option

within ~1.3× of the best-value model · 1.3–2× · >2× · ★ this model is the best-value pick on that task. Top-right = best quadrant. Only tasks where this model qualifies at the 90% bar are plotted.

Per-task breakdown

TaskCategoryQuality (% of best)ConfidenceOverpay
Topic Section Assignment ★Topic Organization & Clustering97%MEDIUMbest value
Social Post Portfolio SelectionRelevance, Classification & Matching91%HIGH1.3x
Topic Sequence OptimizationTopic Organization & Clustering93%HIGH3.2x
Visual Theme Configuration GenerationLong-form Content Generation91%MEDIUM3.2x
Promotional Message GenerationSocial & Promotional Content91%HIGH5.8x
Topic Cluster LabelingTopic Organization & Clustering91%RANKED7.3x
Structured Content SummarizationContent Summarization & Synthesis97%RANKED8.3x
Public Response GenerationSocial & Promotional Content93%HIGH9.1x
Factual Claim RefinementInfrastructure & Utility96%HIGH9.6x
Atomic Fact Claim ExtractionStructured Data & Fact Extraction91%MEDIUM10x
Engagement Opportunity TriageRelevance, Classification & Matching96%RANKED11x
Language IdentificationRelevance, Classification & Matching98%RANKED11x
Content To Image Prompt GenerationInfrastructure & Utility92%RANKED11x
Model Specific Prompt AdaptationInfrastructure & Utility95%RANKED12x
Bench Hypothesis Post GenerationSocial & Promotional Content94%MEDIUM16x
Batch Text TranslationInfrastructure & Utility90%MEDIUM18x

Overpay — how much more you pay by running this model instead of the best-value model that clears the quality bar on that task (marked ★). "16x" means you overpay 16× — the same output for 16× the best-value good-enough option; ★ means this model is that option (no overpayment). Confidence — how sure we are about the quality score (more judgments + more agreement = higher confidence): RANKED many independent judges scored this model's outputs and their agreement is very high (most confident) — HIGH many judges have scored it and they mostly agree (well-pinned) — MEDIUM enough judges have weighed in to publish, but they disagree more than we'd like (treat with a small grain of salt). LOW-confidence cells are hidden everywhere on the site. See the methodology for the exact thresholds.