live wire
▸ThoughtSpot makes Spotter 3 default for new instances; it requires download rights and always shares query data with an LLMThoughtSpot documentation▸MotherDuck adds image classification and object location to prompt_jev(), returning typed SQL outputs with confidence scoresMotherDuck▸BigQuery conversational analytics adds Montréal and Toronto; mixed-region agents still default to globalGoogle Cloud Documentation▸Fabric enables Azure OpenAI Copilot access by default, but OpenAI-subprocessor and cross-region processing/storage switches remain offMicrosoft Learn▸Snowflake Cortex Agents now continue by default when named tools are inaccessible, emitting warnings instead of failing the runSnowflake documentation▸Databricks Genie One can use group memberships to tailor answers, while Unity Catalog remains the access-control boundaryDatabricks documentation▸Fabric data agents can query Power BI models with Read alone; Build and workspace access are unnecessaryMicrosoft Learn▸Fabric data-agent Git configs expose column flags, but selecting a table still includes every columnMicrosoft Learn▸Google Data Agent Kit offers remote Spark kernels, but notebook exports can still move results into local Pandas or ArrowGoogle Cloud Documentation▸Cube says production analytics agents need explicit partial-result states, caller-filtered model search and durable artifactsCube▸Thinking Machines names Bridgewater behind ReViSQL-K2.6, citing 92.97% SC-16 accuracy at 15% of frontier-model costThinking Machines Lab▸AWS says a mismatched data/tool/prompt release cut its Field Sales agent pass rate from 78% to 38% in pre-productionAWS Artificial Intelligence Blog▸NL2SQL Multi-Agent ships evaluation artifacts across BIRD, CSpider, Archer and KDD DataSpace—but no software licenseGitHub▸Snowflake says Cortex Analyst apps keep working, but recommends Cortex Agents; semantic views and verified queries carry overSnowflake Documentation▸ThoughtSpot makes Spotter 3 default for new instances; it requires download rights and always shares query data with an LLMThoughtSpot documentation▸MotherDuck adds image classification and object location to prompt_jev(), returning typed SQL outputs with confidence scoresMotherDuck▸BigQuery conversational analytics adds Montréal and Toronto; mixed-region agents still default to globalGoogle Cloud Documentation▸Fabric enables Azure OpenAI Copilot access by default, but OpenAI-subprocessor and cross-region processing/storage switches remain offMicrosoft Learn▸Snowflake Cortex Agents now continue by default when named tools are inaccessible, emitting warnings instead of failing the runSnowflake documentation▸Databricks Genie One can use group memberships to tailor answers, while Unity Catalog remains the access-control boundaryDatabricks documentation▸Fabric data agents can query Power BI models with Read alone; Build and workspace access are unnecessaryMicrosoft Learn▸Fabric data-agent Git configs expose column flags, but selecting a table still includes every columnMicrosoft Learn▸Google Data Agent Kit offers remote Spark kernels, but notebook exports can still move results into local Pandas or ArrowGoogle Cloud Documentation▸Cube says production analytics agents need explicit partial-result states, caller-filtered model search and durable artifactsCube▸Thinking Machines names Bridgewater behind ReViSQL-K2.6, citing 92.97% SC-16 accuracy at 15% of frontier-model costThinking Machines Lab▸AWS says a mismatched data/tool/prompt release cut its Field Sales agent pass rate from 78% to 38% in pre-productionAWS Artificial Intelligence Blog▸NL2SQL Multi-Agent ships evaluation artifacts across BIRD, CSpider, Archer and KDD DataSpace—but no software licenseGitHub▸Snowflake says Cortex Analyst apps keep working, but recommends Cortex Agents; semantic views and verified queries carry overSnowflake Documentation
nl2sql.ai
benchmarkBenchmark

Genloop leads Spider 2.0-Snow at 96.70%—but the public method trail is thin

The official leaderboard verifies the score; Genloop’s site explains its context product, not enough of the evaluated system to reproduce the result.

Spider 2.0-Snow leaderboard with Genloop narrowly in first place.
Chart: figures from the story
By The Benchmark Desk· Oct 16, 2026the quick take — two AI hosts go live when you do

A narrow lead at the top

Genloop’s Sentinel Agent v2 Pro now holds the top position on the official Spider 2.0-Snow leaderboard with a 96.70% success rate, observed October 16, 2026. The runner-up, Native mini, is listed at 96.53%, so the margin is 0.17 percentage points. The same leaderboard also lists Genloop’s earlier Sentinel Agent v2 at 88.48% and v1.5 at 83.36%. Those are separate submitted systems, not evidence by themselves that a single controlled change produced the gains. Spider 2.0 leaderboard

Spider 2.0-Snow contains 547 Snowflake text-to-SQL tasks. The benchmark asks systems to produce final CSV outputs rather than merely SQL text, and the site says officially uploaded scores follow its validation process. That makes 96.70% a tracked benchmark result, not a local development-set claim. Spider 2.0 benchmark

What the public record does—and does not—show

Genloop’s own site calls the company’s product a context engine for B2B AI and prominently claims the No. 1 Spider 2.0-Snow position. It describes a per-customer “living graph” built across tables, documents and people, served over MCP and API; it also says agents receive cited answers and that expert-approved corrections are rechecked. Genloop

But the public pages opened for this check do not connect those product components to a reproducible Sentinel Agent v2 Pro recipe. They do not state which base model powered the submission, how many model calls or retries it used, what tools were available, what prompt and retrieval policy ran, or how much the run cost. The official leaderboard row likewise names the system and score but exposes no paper or code link for that entry. Spider 2.0 leaderboard Genloop

That distinction matters because the numerical lead is smaller than the disclosure gap. A 0.17-point advantage may be meaningful on 547 enterprise-style tasks, but readers cannot attribute it to Genloop’s graph, an underlying model, search budget, repair loop or another component from the currently public evidence. Nor should the jump from Sentinel Agent v2’s 88.48% to v2 Pro’s 96.70% be read as an ablation: the leaderboard does not identify a controlled variable between those rows. Spider 2.0 leaderboard

The defensible conclusion is therefore precise: Genloop owns the highest officially listed Spider 2.0-Snow score as of October 16. The result establishes rank, but not yet a reproducible explanation for that rank. A method card naming the model, tool budget, retry policy, prompts and execution traces would turn an impressive number into evidence other teams could test.

sources

Filed by The Benchmark Desk. Corrections: desk@nl2sql.ai · Our standards →

comments · 0

    Comments are moderated before they appear. Your email is used once to confirm it is you — never shown, never sold. Corrections and questions get an answer from the desk when we have one.

    more from the desk