DataBench: Frontier Leaderboard for Agentic Analytics | Hex<br>Skip to main content

Last run: August 13, 2026<br>arrow-right-mediumLearn more about DataBench

Models are ranked by blended DataBench score across representative analytics tasks.<br>Hover to compare cost, token usage, and reasoning effort.

DataBench score by average cost per taskModelAverage cost per taskDataBench scoreFable 5 Max$4.785%Fable 5 High$2.671%Fable 5 Medium$2.176%Fable 5 Low$1.572%Opus 5 Max$3.170%Opus 5 XHigh$2.880%Opus 5 High$2.188%Opus 5 Medium$1.483%Opus 5 Low$1.070%GPT-5.6 Sol XHigh$1.275%GPT-5.6 Sol High$0.972%GPT-5.6 Sol Medium$0.771%GPT-5.6 Sol Low$0.565%Opus 4.8 Max$2.874%Opus 4.8 XHigh$1.870%Opus 4.8 High$1.470%Opus 4.8 Medium$1.170%Opus 4.8 Low$0.872%GPT-5.6 Luna XHigh$0.173%GPT-5.6 Luna High$0.164%GPT-5.6 Luna Medium$0.064%GPT-5.6 Luna Low$0.150%GPT-5.6 Terra XHigh$0.468%GPT-5.6 Terra High$0.363%GPT-5.6 Terra Medium$0.357%GPT-5.6 Terra Low$0.270%Sonnet 5 Max$1.461%Sonnet 5 XHigh$0.964%Sonnet 5 High$0.760%Sonnet 5 Medium$0.556%Sonnet 5 Low$0.455%Kimi K2.7 Medium$0.353%Average cost per task

Collections call list<br>I need the January 15 collections call list. Looks like Commerce and Stripe disagree on some subscription statuses— which accounts should we call, and how much MRR is at risk?<br>107 accounts to call, with $139,131/month of MRR at risk and $483,487 of unpaid invoices behind it. The important finding is that the Commerce/Stripe status disagreement is not the call list...

Judging: 100% passModel /Opus 5 (High)<br>Cost /$3.19<br>Input/output tokens /2.4M/54.3K<br>Latency /13m 1s

Acquisition duplicate merges<br>Which OfficeMax duplicates should we merge before the January customer count?<br>...critically, zero OMD identities have ever been linked into the public customer identity registry... Merge the 100 duplicate_candidate accounts. That's the set the acquisition process itself flagged...

Judging: 0% passModel /Fable 5 (Max)<br>Cost /$8.49<br>Input/output tokens /2.3M/94.8K<br>Latency /12m 28s

ModelScorechevron-downCost/TaskTokens/TaskMedian latency/Task1Opus 5 · High88.0%$2.1342K482.4s2Fable 5 · Max85.0%$4.7355.3K406.2s3Opus 5 · Medium83.0%$1.3725.9K190.9s4Opus 5 · XHigh80.0%$2.7759.3K450.5s5Fable 5 · Medium76.0%$2.1017.9K182.9s6GPT-5.6 Sol · XHigh75.0%$1.2219.5K429.5s7Opus 4.8 · Max74.0%$2.8275.7K489.8s8GPT-5.6 Luna · XHigh73.0%$0.0916.5K193.3s9GPT-5.6 Sol · High72.0%$0.9214.7K275s10Opus 4.8 · Low72.0%$0.8415K131.4s11Fable 5 · Low72.0%$1.4811.5K111s12Fable 5 · High71.0%$2.5524.4K188.3s13GPT-5.6 Sol · Medium71.0%$0.7111.1K149.9s14Opus 4.8 · XHigh70.0%$1.7843.7K353s15Opus 5 · Max70.0%$3.0667.6K484.1s16Opus 5 · Low70.0%$0.9717.5K160.9s17Opus 4.8 · High70.0%$1.3828.4K465.1s18Opus 4.8 · Medium70.0%$1.1123K188.5s19GPT-5.6 Terra · Low70.0%$0.206.8K102.3s20GPT-5.6 Terra · XHigh68.0%$0.4015.6K350.5s21GPT-5.6 Sol · Low65.0%$0.467.9K123.6s22GPT-5.6 Luna · High64.0%$0.0713.4K187.3s23GPT-5.6 Luna · Medium64.0%$0.057.3K143.4s24Sonnet 5 · XHigh64.0%$0.8743.5K370.3s25GPT-5.6 Terra · High63.0%$0.2810.4K120.8s26Sonnet 5 · Max61.0%$1.4083.6K453.2s27Sonnet 5 · High60.0%$0.6730.7K213.9s28GPT-5.6 Terra · Medium57.0%$0.258.3K638.6s29Sonnet 5 · Medium56.0%$0.5420.6K462.2s30Sonnet 5 · Low55.0%$0.3513K199.3s31Kimi K2.7 · Medium53.0%$0.2910.4K83.2s32GPT-5.6 Luna · Low50.0%$0.065.9K352.5s

Made with🍩<br>🥟<br>🍺<br>🍰<br>🔮<br>🔒<br>🥖<br>🍷<br>🛌<br>💜<br>🥨<br>🛹<br>🍤<br>🧄<br>🍞<br>🥥<br>🤞<br>🔊<br>🎧

on 🌎.

Company<br>AboutCareersCustomersSolutionsMedia kitNewsroom<br>Platform<br>AI and agentsAgentic notebooksConversational self-serveContext StudioHex CLIExploratory analysisEmbedded analyticsData appsIntegrationsChangelog<br>Resources<br>PricingSwitching to HexEnterpriseDocsBlogEventsTemplatesCompareTrust CenterStatus<br>Connect<br>Contact salesRequest a demoTechnical supportLinkedInX (Twitter)YouTube

© 2026 Hex Technologies Inc.<br>Privacy policyTerms & conditionsModern slavery statementYou have opted out of data tracking