Only 4 of 23 deep-research benchmarks ever changed hands — the rest still sit with their first measured champion

Asked (summary):

Recreate and enrich the benchmarking timeline from the collected research page, with a common timeline, click-on/off players, and a view of who the champion is.

Asked (follow-up):

“Could you use logos/favicons for players in this report?”

The collected page yields 28 champion events across 23 benchmarks, from February 2025 to October 2026, spanning 7 providers. Each point below is a benchmark’s first observed winner or a later record-breaking takeover; scores use each benchmark’s own metric and are not comparable across rows, so there is no single universal champion.

Click a provider chip to hide or show its events; each chip and marker carries the provider’s official-site favicon, with a lettermark fallback if the image cannot load. Valyu appears in the controls but has no dated champion event in this data. Drag the scrubber or press play to watch the current-holder column update as records fall. Month-precision dates are plotted at the first of the month but displayed at month precision. Hover any point for full details.

Only four benchmarks show a takeover in this collection: BrowseComp went OpenAI → Gemini → Parallel, finishing at 94% parallel.ai.

DSQA moved Gemini → You.com at 86.61% in July 2026 about.you.com, and HLE moved Perplexity → Gemini at 46.4% blog.google.

WANDR moved OpenAI → Perplexity at 0.363 soft F1 in July 2026 perplexity.ai.

Current records split across providers: Perplexity holds 6 benchmark rows, Keenable and OpenAI 5 each, Gemini 3, You.com 2, Parallel and Exa 1 each — the latest, NEEDLE at 73%, was set by Keenable on 2026-10-21 keenableai.github.io.

Current record holders

One row per benchmark. 19 of 23 rows are a first recorded holder with no later takeover observed in the collected data.

Benchmark	Current holder	Record score	Date set	Events	History	Source

Method: record progression extracted from the collected research.site timeline page (research.site/table?view=charts) — 28 champion-event rows across 23 benchmarks and 23 latest-holder rows, Feb 2025 to Oct 2026, citing 26 distinct source URLs on 14 independent hosts; the largest single URL supports 2 rows (7.1%). Each score is the benchmark’s own metric (accuracy %, F1, KPR, rubric score), normalized 0–100 only for plotting within a row, never for cross-benchmark comparison. “First measured champion” means first winner observed in this collection, not benchmark inventor or all-time record; month-precision dates are plotted at the first of the month. This is not necessarily the complete world-record history of every benchmark.

This report was generated automatically by Keenable SELECT at a user's request, from publicly available web sources linked herein. Keenable does not review, verify, or endorse its contents and makes no representation as to accuracy, completeness, or timeliness; AI-based extraction may contain errors. Nothing in this report is investment, legal, financial, or other professional advice. All trademarks and referenced content remain the property of their respective owners; no affiliation or endorsement is implied. To report an error, rights concern, or request removal: legal@keenable.ai.