The question effectively asks whether, by the end of May 2026, the third‑ranked frontier AI model (usually meaning a general LLM for chat/agentic work) will be a Google (Gemini) model, and whether Google itself will be viewed as holding the “third best” position in the AI model race. In practice, there are two main ways this is being operationalized in 2026:

1. LMSYS / Arena Chatbot Leaderboard – human‑preference Elo rankings for chat models (often the canonical “who’s best” metric).

2. Prediction markets (especially Polymarket) – markets explicitly titled along these lines, which resolve by reference to LMSYS/Arena rankings.

The competition at the top currently centers on OpenAI (GPT‑5.x / 5.5), Anthropic (Claude Opus / Mythos), and Google (Gemini 3.x), with xAI’s Grok and DeepSeek as emerging contenders.

Key Factors

1. How “third best” is being defined

• Several prediction markets explicitly reference LMSYS/Arena as the resolution source for “best AI model” and “third best AI model” questions, including variants like “Which company has the third best AI model end of May?” and a style‑control version (Style Control On/Off) [1][2].

• One such market states: “This market will resolve according to the company that owns the model that has the third‑highest arena rank … based on the Chatbot Arena LLM Leaderboard” [3].

• Another derivative tracker (ProbSee/Lines) shows live volumes and probabilities for that same “third best AI model end of May” market, listing Anthropic, Google, Baidu, OpenAI, Meta, etc. as outcomes with associated volumes and price‑implied probabilities [4].

So, “third best AI model” in this context almost always means: third‑highest ranked model on LMSYS/Arena at end‑of‑month, mapped back to its owning company.

2. Current rankings and performance of Gemini vs GPT‑5 vs Claude

Benchmarks and composite rankings:

• LM Council benchmarks (March 2026) show Gemini 3.1 Pro Preview at the top of one key aggregate score table (79.6 %), ahead of GPT‑5.5 Pro and Gemini 3 Pro Preview in that specific metric [5].

• BenchLM model page for Gemini 3.1 Pro reports it as very near the frontier across multiple task groups (e.g., top 3 on knowledge/understanding at 94.8 % average) [6].

• Independent comparison pieces put Gemini 3.1 Pro near – but often just behind – GPT‑5.5 and Claude Opus 4.7 on various axes:

• GPT‑5.5 winning on agentic/coding/computer‑use benchmarks like Terminal‑Bench 2.0 [7].

• Claude Opus 4.7 leading especially on reasoned coding and tool‑heavy workflows [7].

• Gemini 3.1 Pro leading reasoning benchmarks such as ARC‑AGI‑2 and some multimodal understanding suites [8][9].

Arena/LMSYS context:

• Multiple secondary write‑ups summarising LMSYS data describe Gemini 3.1 Pro:

• Early 2026: “top 3 across Text and Vision Arena” and #6 in Code Arena at launch [10].

• March 2026: overall Elo around 1476 and usually behind GPT‑5.4 and/or Claude Opus 4.6 in text [11].

• An April 2026 article notes Claude Opus 4.6 taking #1 in text, with Gemini 3.1 Pro just a few Elo below [12].

• A March‑2026 leaderboard history piece notes the top Arena Elo rising to ~1501 (Claude Opus 4.6 “thinking”) with Gemini 3.1 Pro and GPT‑5.x clustered just below [13].

In short: Gemini is consistently in the top tier but not clearly entrenched at exactly #3, and the precise ordering among GPT‑5.5, Claude Opus 4.7, Grok 4.x and Gemini 3.x shifts by benchmark and by Arena category.

3. Prediction market odds about the “third best AI model”

We have several relevant markets and summaries:

• A Polymarket event “Which company has the third best AI model end of May?” explicitly uses LMSYS/Arena as its resolution source. A summary snippet states:

  “The current frontrunner for ‘Which company has the third best AI model end of May?’ is ‘Anthropic’ at 67 %, meaning the market assigns a 67 % chance to that outcome.” [14]

• A different aggregator states:

  “Trader consensus on Polymarket gives Anthropic a 66 % implied probability of holding the third‑best large language model on the LMSYS Chatbot Arena leaderboard …” [15].

• For earlier horizons (April 2026), there was a period where Google was near‑certain in another “third best” market:

• A Perplexity‑style finance summary reports: “Google’s near‑certain 100 % probability for third‑best AI model reflects the April 2026 leaderboard reality where Gemini 3.1 Pro …” (the snippet implies that for April, traders believed Gemini was locked in at third) [16].

• For May, however, these live odds have moved away from Google, with multiple snippets consistently putting Anthropic as the leading candidate for third best at the end of May, while Google appears as a significant but secondary probability mass (around the low‑30s) [15][17].

These markets are not proof of the future, but they are forward‑looking, money‑weighted consensus about what LMSYS will show at month‑end. Right now they do not expect Google to be the most likely holder of the third‑best slot.

4. Impact of upcoming Gemini 3.2 releases

• A mid‑May article notes that Google will unveil Gemini 3.2 Flash at I/O 2026 on May 20, claiming 92 % “performance efficiency” vs GPT‑5.5 and lower costs [18].

• Another leak‑type report suggests Gemini 3.2 Ultra will be highlighted at I/O 2026 with substantial benchmark gains [18].

• However, a Manifold‑style market analysis points out that as of May 12, 2026, Gemini 3.2 is not yet officially released, and uncertainty remains about whether it will be on LMSYS in time for end‑of‑May resolution [18].

Even if 3.2 is unveiled at I/O, there are two practical constraints:

1. Deployment lag to LMSYS/Arena – Historically there is a delay between a model’s announcement and its incorporation and stabilization on LMSYS Chatbot Arena.

2. Data‑collection window – Arena Elo requires a significant volume of head‑to‑head votes; a model launched in mid‑to‑late May might not accumulate stable Elo in time to displace top incumbents like GPT‑5.5 or Claude Opus 4.7 before the May 31 snapshot.

So while Gemini 3.2 could shock the ecosystem, markets that explicitly track this risk are still assigning Anthropic, not Google, the highest probability of exactly third place.

5. Cross‑leaderboard and methodology caveats

• Different leaderboards emphasize different objectives:

• LMSYS/Arena – human preference in chat.

• Artificial Analysis, BenchLM, AiZolo, LM Council, Vellum – composite indices blending pure benchmarks, cost, speed, and sometimes human preference [19][20][5].

• One meta‑leaderboard article stresses that a model can be top‑3 on one scoreboard and only top‑10 on another, and both statements can be “correct” because they measure different things [21].

• Because both prediction markets and multiple “best AI” debates explicitly reference LMSYS/Arena as tie‑breaker or resolution source [22][23], the most relevant notion of “third best” for this question is: Arena ranking, not internal Google metrics or cost‑adjusted value rankings.

Taken together, the relevant frame is: “At the end of May 2026, on the LMSYS Chatbot Arena leaderboard (or equivalent Arena text leaderboard), will the #3 model belong to Google?”

Synthesis and Direct Answer

Based on:

• Current Arena‑aligned prediction markets, which consistently show Anthropic, not Google, as the favorite to end up with the third‑best model at the end of May 2026, with Anthropic around 65–70 % implied probability and Google in the 20–35 % range [15][14][17].

• The fact that Gemini 3.1 Pro is already a top‑tier model but is generally depicted as competing for top‑3 overall, often fighting for 2nd or 3rd with Claude and GPT‑5.5, rather than being securely at exactly #3.

• The timing and uncertainty around Gemini 3.2 reaching LMSYS and accumulating enough data to alter the ranking before the May 31 cutoff [18].

The most defensible forecast is:

It is unlikely that Google will be the owner of the exactly third‑best AI model at the end of May 2026 under the LMSYS/Arena definition, although it does have a non‑trivial chance (roughly 25–35 %). In expectation, Anthropic is more likely to occupy the third‑place slot, with Google more likely occupying either second or fourth than exactly third.

In other words: Google will almost certainly still be in the top cluster (top 3–4 models), but the best available data and market consensus say it is not the most probable candidate to finish exactly third.