Fronset Benchmark

Thinking Machines Inkling Small at a glance

Good enough on 26/54 tasks at the Balanced step. Best value on 4 tasks. Best fit for: Long-form Content Generation, Relevance, Classification & Matching, Social & Promotional Content.

Provider
OpenRouter
Model name
thinkingmachines/inkling-small
What the steps mean
Cost mode:
Qualifies on
26 / 54 tasks (at the Balanced step)
Best value on
4 tasks

Cost vs quality across all tasks

0%25%50%75%100%678910Quality score (6–10)Cost-efficiency vs best value (1.0 = best value)Factual Claim Refinement — quality 8.31, 7.7x the cost of the best-value good-enough optionAtomic Fact Claim Extraction — quality 7.34, 13.2x the cost of the best-value good-enough optionResearch Query Generation — quality 7.26, 10.6x the cost of the best-value good-enough optionMulti Perspective Decision Synthesis — quality 8.45, 7.5x the cost of the best-value good-enough optionBench Hypothesis Post Generation — quality 8.24, 11.6x the cost of the best-value good-enough optionEngagement Opportunity Triage — quality 8.78, 21.9x the cost of the best-value good-enough optionPublic Response Generation — quality 8.06, 31.5x the cost of the best-value good-enough optionTopic Cluster Labeling — quality 8.28, 40.5x the cost of the best-value good-enough optionSocial Post Portfolio Selection — quality 8.69, 19.2x the cost of the best-value good-enough optionTopic Section Assignment — quality 8.01, 8.2x the cost of the best-value good-enough optionReport Outline Generation — quality 9.24, this model IS the best-value good-enough option★Content Set Relevance Scoring — quality 6.74, 2.9x the cost of the best-value good-enough optionResearch Region Identification — quality 8.48, this model IS the best-value good-enough option★Profile Pool Matching — quality 7.86, 12.5x the cost of the best-value good-enough optionContent To Image Prompt Generation — quality 8.68, 11.9x the cost of the best-value good-enough optionModel Specific Prompt Adaptation — quality 8.09, 5.9x the cost of the best-value good-enough optionVoice Profile Generation — quality 8.65, 6.0x the cost of the best-value good-enough optionVisual Theme Configuration Generation — quality 8.76, 8.4x the cost of the best-value good-enough optionStructured Content Summarization — quality 8.23, 8.4x the cost of the best-value good-enough optionPublication Title Package Generation — quality 8.45, 18.9x the cost of the best-value good-enough optionLanguage Identification — quality 9.77, 9.9x the cost of the best-value good-enough optionGenerated Content Relevance Scoring — quality 7.91, 17.9x the cost of the best-value good-enough optionPromotional Message Generation — quality 8.15, 15.6x the cost of the best-value good-enough optionBatch Text Translation — quality 8.48, 12.9x the cost of the best-value good-enough optionBench Article Hypothesis Generation — quality 8.39, this model IS the best-value good-enough option★Content Domain Suggestion — quality 7.37, this model IS 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 Balanced step are plotted.

Per-task breakdown

TaskCategoryQuality (% of best)ConfidenceOverpay
Bench Article Hypothesis Generation ★Evidence-Grounded Hypothesis Generation96%MEDIUMbest value
Report Outline Generation ★Long-form Content Generation98%RANKEDbest value
Research Region Identification ★Structured Data & Fact Extraction96%HIGHbest value
Content Domain Suggestion ★ bestRelevance, Classification & Matching100%MEDIUMbest value
Content Set Relevance ScoringRelevance, Classification & Matching99%MEDIUM2.9x
Model Specific Prompt AdaptationInfrastructure & Utility91%MEDIUM5.9x
Voice Profile GenerationLong-form Content Generation90%MEDIUM6x
Multi Perspective Decision SynthesisFinancial Analysis & Trading Decisions96%HIGH7.5x
Factual Claim RefinementInfrastructure & Utility96%HIGH7.7x
Topic Section AssignmentTopic Organization & Clustering98%MEDIUM8.2x
Structured Content SummarizationContent Summarization & Synthesis93%MEDIUM8.4x
Visual Theme Configuration GenerationLong-form Content Generation99%HIGH8.4x
Language IdentificationRelevance, Classification & Matching98%RANKED9.9x
Research Query GenerationInfrastructure & Utility92%MEDIUM11x
Bench Hypothesis Post GenerationSocial & Promotional Content93%MEDIUM12x
Content To Image Prompt GenerationInfrastructure & Utility97%MEDIUM12x
Profile Pool MatchingRelevance, Classification & Matching90%MEDIUM13x
Batch Text TranslationInfrastructure & Utility92%MEDIUM13x
Atomic Fact Claim ExtractionStructured Data & Fact Extraction95%MEDIUM13x
Promotional Message GenerationSocial & Promotional Content94%MEDIUM16x
Generated Content Relevance ScoringRelevance, Classification & Matching91%MEDIUM18x
Publication Title Package GenerationContent Summarization & Synthesis92%RANKED19x
Social Post Portfolio SelectionRelevance, Classification & Matching100%HIGH19x
Engagement Opportunity TriageRelevance, Classification & Matching99%RANKED22x
Public Response GenerationSocial & Promotional Content95%MEDIUM31x
Topic Cluster LabelingTopic Organization & Clustering93%MEDIUM41x

Overpay — how much more you pay by running this model instead of the best-value model that meets the quality step 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.