Two scoreboards claim to rank every major AI model. Arena, once called LMArena, has run since 202311. Artificial Analysis publishes a second count, the Intelligence Index3. Product leaders check both before they sign a contract.
AI Lately counted who held first place on each board for the last year. Google and Anthropic are the only two labs that ever got there. OpenAI led one board for twenty weeks and touched zero on the other123.
Two Scoreboards, Two Different Tests
Arena asks real people to vote. A visitor types one prompt. Two anonymous models answer it. The visitor picks the better reply, and Arena turns millions of those picks into a single Elo-style score11.
Elo is a rating system borrowed from chess. A win against a strong opponent raises a score more than a win against a weak one.
Ion Stoica helped start that project at UC Berkeley in 202311. Arena grew fast. It raised $200 million in October and now carries a $3.1 billion price tag10.
Artificial Analysis skips human votes. It gives every model the same written tests and adds up the score3. The company rebuilt that test six times in the past year, moving from v3 to v5456.
Each new version adds harder questions, so old scores stop comparing to new ones. Version resets explain why this count tracks dates and labs across each rebuild, ahead of raw scores.
Counting Who Held #1
AI Lately picked one start date both boards agree on: Nov. 18, 2025, the day Gemini 3 Pro reached #1 on each12. From there, AI Lately logged every dated change either board posted through Nov. 4, 2026, a window of fifty weeks.
The method is simple on purpose. A reader can repeat it on either site. Log the date a new model takes first place, then log the lab behind it. Add up the days until the next change.
Arena's own Elo history, mirrored by the tracking site BenchLM, shows three stretches at the top in that window2. Google's Gemini 3 Pro led first. Anthropic held the top spot next, under two different Claude models, from February to September. Google then retook first place with Gemini 4 Argon, a lead it still holds today behind Gemini 4 Carbon1.
6 Dated Changes Split Two Boards in a Year
Dated changes in which model led Arena's Text Arena and Artificial Analysis's rankings, Nov. 2025 to Oct. 2026
Each pin is one date a board recorded a new #1 model. Magenta marks Anthropic, blue marks Google, and gray marks OpenAI's one turn at the top.
Two of the six dates come from Arena's own monthly history and carry month-level precision alone.
Sources: Arena [1]; BenchLM [2]; Artificial Analysis [3]; OfficeChai [4]; Artificial Analysis [5]; NextBigFuture [7]. Chart by Kelly Dennis.
The numbers behind this chart
| Date | Event |
|---|---|
| Nov. 18, 2025 | Gemini 3 Pro leads both. Google; leads Arena and the AA board the same day |
| Jan. 6, 2026 | GPT-5.2 leads AA's board. OpenAI; Artificial Analysis rebuilt its scoring this day |
| Feb. 2026 | Opus 4.6 leads Arena. Anthropic; month-level date from Arena's own history |
| May 28, 2026 | Opus 4.8 leads AA board. Anthropic |
| Sept. 2026 | Argon leads Arena. Gemini 4 Argon, Google; month-level date from Arena's own history |
| Oct. 28, 2026 | Carbon leads AA board. Gemini 4 Carbon, Google |
Run the same count on the Artificial Analysis Index, and the order changes completely. Google's Gemini 3 Pro led first there too. OpenAI's GPT-5.2 then took over when the index rebuilt its scoring on Jan. 64.
Anthropic's Claude Opus 4.8 claimed the top spot on May 28. It held first place through two more Anthropic models, until Google's Gemini 4 Carbon overtook it on Oct. 2856.
The Lab Still Waiting for #1
Add up the weeks, and Anthropic leads both counts. It held Arena's top spot for thirty of the last fifty weeks and the Artificial Analysis Index for twenty-two123. Google held Arena for twenty weeks and the Index for eight. Eight weeks, in a year of trying.
OpenAI's count tells a sharper story. GPT-5.2 held the Artificial Analysis Index for twenty weeks, the same total Google managed on Arena4. On Arena's Text Arena itself, OpenAI's count stayed at 0 the entire year.
Anthropic Topped Arena's Text Leaderboard 30 of the Last 50 Weeks
Weeks each lab held #1 on Arena's Text Arena and Artificial Analysis's rankings, Nov. 18, 2025 to Nov. 4, 2026
Each bar is one leaderboard. The segments show how many of the last 50 weeks each lab held its #1 spot, and Arena's top spot went only to Google and Anthropic across the whole window.
Arena's Text Arena ranks models by human votes between two chatbots. The Artificial Analysis Index ranks them by a weighted composite of written tests. The boards measure different things.
Sources: Arena [1]; BenchLM [2]; Artificial Analysis [3]; OfficeChai [4]; Artificial Analysis [5]; Artificial Analysis [6]. Chart by Kelly Dennis.
The numbers behind this chart
| Item | Anthropic | OpenAI | |
|---|---|---|---|
| Arena's Text Arena | 20 wk | 30 wk | 0 wk |
| Artificial Analysis Index | 8 wk | 22 wk | 20 wk |
| Whole | 50 wk | ||
Five more labs missed first place on either board: xAI, Meta, Moonshot, DeepSeek and MiniMax. xAI's Grok 4.1 came closest, holding Arena's top spot for one day before Gemini 3 Pro bumped it off7.
Arena.ai@arenaToday, we're announcing our Series B: $200M at a $3.1B valuation, and the release of Arena's Alignment Index.
AI is advancing faster than our ability to evaluate it. The world needs a neutral third party to measure how safe and aligned AI actually is once it's in the hands of… https://t.co/SkKhEQ9swC pic.twitter.com/vUc50MsiyP
Does the Top Spot Decide the Sale
Micah Hill-Smith built the Artificial Analysis Index and still defends it. "I feel comfortable asserting that it has consistently been, and will continue to be, the best single metric for model intelligence," he told Trending Topics8.
Arvind Narayanan reads the same leaderboards and reaches a different number. He directs Princeton's Center for Information Technology Policy. With co-author Sayash Kapoor, he studies how teams actually choose AI systems. Their own paper argues that "downstream developers should keep in mind that HumanEval or any other standardized benchmark is nothing more than a rough proxy for the specific tasks that arise in a particular downstream application"9.
Both read the same two boards and reach opposite advice. Hill-Smith runs one of those boards himself, so a trusted score sells his own product too. Narayanan studies how buyers behave on the job, apart from how vendors market, and his research follows the invoice ahead of the rank89.
Arena's own history backs part of that skepticism. A Meta model once scored higher on Arena than the public version buyers could download. Researchers later found hundreds of manipulated votes on the board11. Moonshot's Kimi K3.1 Pro reached fourth place on the Artificial Analysis Index in October while charging far less than the leaders above it6.
What to do
Check both boards before you buy. Rank alone misleads. A model's rank tells a buyer who answered more test questions correctly, or who more strangers preferred in a blind vote. Both boards skip price, uptime, and how a model handles your own data.
Ask a vendor for its own numbers on those three things. A model two spots down the list, at half the price, often wins the actual contract.
Weigh reliability data a vendor rarely volunteers: how often its model stalls, times out, or needs a second try on your own workload.
By the numbers
- Fifty weeks: the span AI Lately counted, from Nov. 18, 2025 to Nov. 4, 202612.
- Thirty weeks: how long Anthropic held Arena's Text Arena top spot12.
- Twenty-two weeks: Anthropic's run atop the Artificial Analysis Intelligence Index3456.
- Twenty weeks: GPT-5.2's run atop the Artificial Analysis Index, matching Google's total on Arena4.
- $3.1 billion: Arena's valuation after its October Series B10.
- One day: how long xAI's Grok 4.1 held Arena's top spot before Google's Gemini 3 Pro took it7.
What to watch
Watch whether Google's Gemini 4 Carbon holds the Artificial Analysis Index past the next version change. The rebuilt index has unseated every leader so far.
Moonshot and DeepSeek are worth watching too. Both have cracked the top five on price alone. A cheaper model that keeps climbing could force both boards to weigh cost the way they now weigh test scores.
Sources
- Arena, "Arena Leaderboard: Official AI Model Rankings & Benchmarks," Arena, Nov. 4, 2026, https://arena.ai/leaderboard/text
- BenchLM, "LLM Leaderboard History," BenchLM, Oct. 23, 2026, https://benchlm.ai/llm-leaderboard-history
- Artificial Analysis, "Comparison of AI Models across Intelligence, Performance, and Price," Artificial Analysis, Nov. 4, 2026, https://artificialanalysis.ai/models
- OfficeChai Team, "Artificial Analysis Rejigs Intelligence Index, GPT 5.2 (xHigh) Takes Top Spot," OfficeChai, Jan. 6, 2026, https://officechai.com/ai/artificial-analysis-rejigs-intelligence-index-gpt-5-2-xhigh-takes-top-spot/
- Artificial Analysis, "Claude Opus 4.8 - The new #1 AI model," Artificial Analysis, May 28, 2026, https://artificialanalysis.ai/articles/claude-opus-4-8-analysis-and-benchmarks
- Artificial Analysis, "Kimi K3.1 Pro launches at 4th on the Intelligence Index and will lead open weights models," Artificial Analysis, Oct. 23, 2026, https://artificialanalysis.ai/changelog
- NextBigFuture Staff, "xAI releases Grok 4.1 and it tops the LMArena leaderboard," NextBigFuture, Nov. 17, 2025, https://nextbigfuture.com/2025/11/xai-releases-grok-4-1-and-it-tops-the-lmarena-leaderboard.html
- Jakob Steinschaden, "How GPT-6 Became No. 1: Artificial Analysis CEO Explains The Much-Debated Index Updates," Trending Topics, Sept. 14, 2026, https://www.trendingtopics.eu/gpt-6-artificial-analysis/
- Sayash Kapoor and Arvind Narayanan, "AI leaderboards are no longer useful. It's time to switch to Pareto curves.," AI as Normal Technology, April 30, 2024, https://www.normaltech.ai/p/ai-leaderboards-are-no-longer-useful
- Arena, "Arena Raises $200M Series B at $3.1B Valuation," Arena, Oct. 8, 2026, https://arena.ai/blog/series-b
- Wikipedia contributors, "LMArena," Wikipedia, Nov. 4, 2026, https://en.wikipedia.org/wiki/LMArena
- 9to5Google Staff, "Gemini 3 Pro Launch," 9to5Google, Nov. 18, 2025, https://9to5google.com/2025/11/18/gemini-3-launch/
