The Right-Sized LLM Benchmark Cheapest LLM that's good enough.

Gemini 3.8 Flash at a glance

Good enough on 15/52 tasks at the 90% bar. Best value on 0 tasks. Best fit for: Infrastructure & Utility, Long-form Content Generation, Topic Organization & Clustering. Doesn't qualify on any: Financi

Provider
Gemini
Model name
gemini-3.8-flash
Cost mode:
Qualifies on
15 / 52 tasks (at 90% bar)
Best value on
0 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)Report Executive Summary Generation — quality 8.15, 1.5x the cost of the best-value good-enough optionReference Preserving Analytical Writing — quality 8.48, 5.1x the cost of the best-value good-enough optionTopic Section Assignment — quality 8.16, 9.3x the cost of the best-value good-enough optionResearch Query Validation — quality 8.75, 5.3x the cost of the best-value good-enough optionPublication Title Package Generation — quality 8.45, 1.6x the cost of the best-value good-enough optionEngagement Opportunity Triage — quality 8.53, 6.7x the cost of the best-value good-enough optionLanguage Identification — quality 9.81, 6.4x the cost of the best-value good-enough optionSocial Post Portfolio Selection — quality 8.72, 16.3x the cost of the best-value good-enough optionVoice Profile Generation — quality 9.34, 4.5x the cost of the best-value good-enough optionBatch Text Translation — quality 9.01, 10.7x the cost of the best-value good-enough optionTopic Taxonomy Matching — quality 8.44, 9.6x the cost of the best-value good-enough optionPromotional Message Generation — quality 8.09, 4.9x the cost of the best-value good-enough optionTopic Cluster Labeling — quality 8.24, 1.8x the cost of the best-value good-enough optionResearch Query Generation — quality 7.89, 7.8x the cost of the best-value good-enough optionPersona Policy Eligibility Check — quality 8.72, 7.2x 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
Report Executive Summary GenerationContent Summarization & Synthesis94%MEDIUM1.5x
Publication Title Package GenerationContent Summarization & Synthesis92%HIGH1.6x
Topic Cluster LabelingTopic Organization & Clustering92%HIGH1.8x
Voice Profile GenerationLong-form Content Generation95%HIGH4.5x
Promotional Message GenerationSocial & Promotional Content93%RANKED4.9x
Reference Preserving Analytical WritingLong-form Content Generation92%MEDIUM5.1x
Research Query Validation bestInfrastructure & Utility100%HIGH5.3x
Language IdentificationRelevance, Classification & Matching98%MEDIUM6.4x
Engagement Opportunity TriageRelevance, Classification & Matching96%HIGH6.7x
Persona Policy Eligibility CheckRelevance, Classification & Matching95%HIGH7.2x
Research Query GenerationInfrastructure & Utility97%HIGH7.8x
Topic Section AssignmentTopic Organization & Clustering94%MEDIUM9.3x
Topic Taxonomy Matching bestRelevance, Classification & Matching100%MEDIUM9.6x
Batch Text TranslationInfrastructure & Utility96%HIGH11x
Social Post Portfolio SelectionRelevance, Classification & Matching99%HIGH16x

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.