Search and agent BI whose Spotter analyst scored 82.8% pass@1 on the BIRD text-to-SQL benchmark in ThoughtSpot's own published five-engine test.
Buyers who refuse to sign a natural-language analytics contract without a number, because ThoughtSpot is one of very few vendors in this category that publishes one: Spotter at 82.8% pass@1 on BIRD dev_20251106 across 1,533 questions and 11 databases, with per-difficulty breakdowns and confidence intervals. It also fits a team that wants both billing modes on the table, since per-user and per-credit prices are both published. Realistically it suits an organisation with modelling capacity, because the docs state Spotter's success depends on the data model - correct synonyms, keywords and joins - and recommend a phased rollout starting with 20 to 50 coached questions.
The 82.8% is a vendor-run study and ThoughtSpot says so in the fine print: competitors are anonymised as categories rather than named, BIRD's official scorer could not run end-to-end because it executes SQL against SQLite and not every engine emits portable SQL, and annotation problems were found in 52.8% of reviewed BIRD Mini-Dev examples. The margins that matter most are the softest - the 3.9-point challenging-band lead over the warehouse-native assistant has not cleared a paired per-question test, and the simple and moderate bands were ties. Plan limits bite on the per-user path: Essentials runs 5 to 50 users with up to 25M rows, Pro caps at 250M rows, and per-user Pro includes only 25 Spotter AI Agent queries per user per month, with unlimited LLM tokens reserved for the per-credit plan. Column sets cannot be used to coach Spotter, and 'why' change-analysis questions require Spotter 3 plus admin enablement.
Pros
Cons
Spotter scored 82.8% pass@1 on the BIRD text-to-SQL benchmark, development split dev_20251106, across 1,533 questions and 11 databases, with one attempt per question. ThoughtSpot reports a frontier LLM plus SQL at 78.4% and a coding agent plus SQL at 72.5%. On the 231 challenging questions Spotter scored 70.6%. The overall 95% confidence interval is roughly plus or minus 2 points.
Read it as a vendor-run study with unusually honest caveats. ThoughtSpot anonymises competitors as categories, states BIRD's official scorer could not run end-to-end because it executes SQL against SQLite, and reports annotation problems in 52.8% of reviewed BIRD Mini-Dev examples. It also flags that the 3.9-point challenging-band lead has not cleared a paired per-question test and that simple and moderate margins were ties.
Through the data model and coaching. The documentation states Spotter's success depends on correct synonyms, keywords, joins and sound structure, and that the model should use terms the team actually uses. Spotter must be enabled at platform, model and user-privilege layers, learns from edits users make to its interpretations, and can be trained with reference questions - the docs suggest starting with 20 to 50 coached questions.
On the per-user path, Essentials starts as low as $25 per user per month and Pro as low as $50, billed annually, with per-user Pro including 25 Spotter AI Agent queries per user per month. On the consumption path, Pro starts as low as $0.10 per credit and lists Spotter with unlimited LLM tokens. ThoughtSpot states it does not meter or charge for LLM tokens.
Facts sourced from the vendor’s own site, last verified 2026-10-10.
The Weekly Source
One email a week. New listings, repriced tools, and the ones we would skip - including this one if its pricing changes.
