Fronset Benchmark

Kimi K2.6 (archived)

Provider
Moonshot AI
Model name
kimi-k2.6
Last benchmarked
Qualified on
29 / 51 tasks (at the 90% bar)
Best value on
0 tasks

Good enough on 29/51 tasks at the 90% bar. Best value on 0 tasks. Best fit for: Structured Data & Fact Extraction, Financial Analysis & Trading Decisions, Social & Promotional Content, Content Summarization & Synthesis.

Cost vs quality across all tasks, as of 15 August 2026

0%25%50%75%100%78910Quality score (7–10)Cost-efficiency vs best value (1.0 = best value)Engagement Triage — quality 8.65, 51.6x the cost of the best-value good-enough optionAuthor Matching — quality 8.39, 28.8x the cost of the best-value good-enough optionOnboarding Chapter Prompt Adaptation — quality 8.92, 25.2x the cost of the best-value good-enough optionContent Summarization — quality 8.57, 15.7x the cost of the best-value good-enough optionGeographic Region Identification — quality 9.39, 31.1x the cost of the best-value good-enough optionImage Prompt Generation — quality 8.50, 26.4x the cost of the best-value good-enough optionSEC Filing Analysis — quality 8.25, 11.3x the cost of the best-value good-enough optionReddit Post Generation — quality 8.33, 4.6x the cost of the best-value good-enough optionX Post Selection — quality 8.68, 15.0x the cost of the best-value good-enough optionInvestment Panel Voting — quality 8.62, 17.7x the cost of the best-value good-enough optionSEC S-1 Chunk Analysis — quality 8.81, 39.2x the cost of the best-value good-enough optionTopic Grouping and Client Matching — quality 7.97, 16.3x the cost of the best-value good-enough optionPublication Title Generation — quality 8.89, 26.1x the cost of the best-value good-enough optionAuthor Living-Person Safety Check — quality 8.87, 20.9x the cost of the best-value good-enough optionLanguage Detection — quality 9.90, 127.4x the cost of the best-value good-enough optionResearch Query Generation — quality 8.45, 24.7x the cost of the best-value good-enough optionSubstack Newsletter — quality 9.16, 26.5x the cost of the best-value good-enough optionTrading Recommendation — quality 8.61, 7.7x the cost of the best-value good-enough optionS-1 TOC Extraction — quality 8.74, 18.7x the cost of the best-value good-enough optionEngagement Reply Draft — quality 8.05, 63.4x the cost of the best-value good-enough optionTopic Report Relevance Scoring — quality 8.47, 14.8x the cost of the best-value good-enough optionStructured Output Extraction — quality 9.71, 68.9x the cost of the best-value good-enough optionPrompt Adaptation — quality 8.76, 48.8x the cost of the best-value good-enough optionTopic Cluster Naming — quality 8.70, 24.0x the cost of the best-value good-enough optionClaim-Referenced Analyst Writing — quality 8.79, 15.1x the cost of the best-value good-enough optionSocial Post Promotion — quality 8.26, 16.2x the cost of the best-value good-enough optionExecutive Summary Generation — quality 7.67, 42.8x the cost of the best-value good-enough optionActivity Feed Blurb Generation — quality 8.66, 53.7x the cost of the best-value good-enough optionTopic-to-Section Assignment — quality 8.11, 16.3x 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, as of 15 August 2026

TaskCategoryQuality (% of best)ConfidenceOverpay
Reddit Post GenerationSocial & Promotional Content94%HIGH4.6x
Trading RecommendationFinancial Analysis & Trading Decisions97%RANKED7.7x
SEC Filing AnalysisFinancial Analysis & Trading Decisions92%MEDIUM11x
Topic Report Relevance ScoringRelevance, Classification & Matching91%HIGH15x
X Post SelectionRelevance, Classification & Matching98%RANKED15x
Claim-Referenced Analyst WritingLong-form Content Generation95%MEDIUM15x
Content SummarizationContent Summarization & Synthesis92%RANKED16x
Social Post PromotionSocial & Promotional Content93%HIGH16x
Topic Grouping and Client MatchingRelevance, Classification & Matching94%HIGH16x
Topic-to-Section AssignmentTopic Organization & Clustering92%MEDIUM16x
Investment Panel VotingFinancial Analysis & Trading Decisions95%RANKED18x
S-1 TOC ExtractionStructured Data & Fact Extraction93%MEDIUM19x
Author Living-Person Safety CheckRelevance, Classification & Matching96%MEDIUM21x
Topic Cluster NamingTopic Organization & Clustering97%RANKED24x
Research Query GenerationInfrastructure & Utility99%HIGH25x
Onboarding Chapter Prompt AdaptationInfrastructure & Utility99%RANKED25x
Publication Title GenerationContent Summarization & Synthesis98%RANKED26x
Image Prompt GenerationInfrastructure & Utility91%RANKED26x
Substack NewsletterLong-form Content Generation97%RANKED26x
Author MatchingRelevance, Classification & Matching99%HIGH29x
Geographic Region Identification bestStructured Data & Fact Extraction100%RANKED31x
SEC S-1 Chunk AnalysisFinancial Analysis & Trading Decisions96%RANKED39x
Executive Summary GenerationContent Summarization & Synthesis97%MEDIUM43x
Prompt AdaptationInfrastructure & Utility98%HIGH49x
Engagement TriageRelevance, Classification & Matching94%RANKED52x
Activity Feed Blurb GenerationSocial & Promotional Content99%RANKED54x
Engagement Reply DraftSocial & Promotional Content95%HIGH63x
Structured Output ExtractionStructured Data & Fact Extraction99%RANKED69x
Language DetectionRelevance, Classification & Matching98%RANKED127x

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.