Model

Anthropic's Claude and OpenAI's GPT-6.2 each top a 3D leaderboard during Anthropic's IPO week

Anthropic, which is expected to price its Nasdaq IPO tonight, holds the crown on one human-vote 3D leaderboard and OpenAI holds the other. Spline, the vendor running the first board, pays its voters in its own AI credits and may hand their votes to research partners.

Vincent JiangVincent Jiangthree-minute read
Share
Dario Amodei, chief executive of Anthropic, speaking on stage at TechCrunch Disrupt in San Francisco
Dario Amodei, chief executive of Anthropic, on stage at TechCrunch Disrupt in San Francisco in September 2023. Anthropic is expected to price its Nasdaq IPO on the evening of November 18, 2026.

On November 3, 2026 Spline, a private browser-based 3D design tool for interactive web scenes, opened Spline Arena to anyone with a free account 1. Two anonymous models each build a scene from the same prompt, users orbit both finished scenes and vote blind, and a Bradley-Terry statistical fit over every head-to-head vote sets the rating 1. Two weeks in, 14 models are ranked 2.

Anthropic's$350B — Anthropic, private, latest valuation $350B Claude Fable 5.2 leads at 1649, winning 62% of its battles at $5.63 and 14 minutes 43 seconds a scene; Google's Gemini 4 Carbon is second at 1620 for $1.60 a scene 2. OpenAI's$1.18T — OpenAI, private, latest valuation $1.18T GPT-6 Astra sits sixth at 1517, winning 44% 2.

Claude Fable 5.2 tops Spline Arena at 1649; OpenAI's best sits sixth at 1517

Anthropic and Google hold the top; OpenAI's best entry is 132 points back

05001,0001,5002,000Claude Fable 5.21,64962% wins, $5.63 a sceneGemini 4 Carbon1,620Gemini 4 Argon1,617Gemini 4 Hydrogen1,598Claude Opus 5.51,593GPT-6 Astra1,517GPT-6.2 not yet on this boardGPT-6.1 Sol1,490DeepSeek V4 Pro1,313
Data
Value
Claude Fable 5.21,649
Gemini 4 Carbon1,620
Gemini 4 Argon1,617
Gemini 4 Hydrogen1,598
Claude Opus 5.51,593
GPT-6 Astra1,517
GPT-6.1 Sol1,490
DeepSeek V4 Pro1,313
Bradley-Terry ratings from blind head-to-head user votes with bootstrapped 95% confidence intervals; 14 models ranked, top eight shown, as of November 18, 2026. Source: Spline Arena rankings.2

GPT-6.2 Astra flips the rival board

On November 17 OpenAI released GPT-6.2 Astra and Sol, and Artificial Analysis, an independent benchmarking firm, put Astra first on its own 3D board, AA-3D, at 1,318 Elo 3. Elo is a chess-style rating that grows with head-to-head wins; human judges picked Astra's scenes in 139 of 152 blind comparisons 3. Claude Fable 5.2 came second at 1,271, Sol at 1,259 3.

AA-3D has run since October 12 with a recruited panel judging 55 tasks in which models write a working three.js web app, browser 3D code, rather than build inside a vendor's editor 34. Spline Arena has not added the 6.2 models, so the board Anthropic leads has not yet scored OpenAI's new champion 2.

Same models, opposite verdicts: Astra leads AA-3D while Fable 5.2 leads Spline Arena

  • Spline Arena rating
  • AA-3D Elo
05001,0001,5002,000Claude Fable 5.2Gemini 4 CarbonGPT-6 AstraGPT-6.1 Sol
Data
Spline Arena ratingAA-3D Elo
Claude Fable 5.21,6491,271
Gemini 4 Carbon1,620—
GPT-6 Astra1,5171,318
GPT-6.1 Sol1,4901,259
Elo-style ratings from two boards with separate voter pools and separate scales, so levels are not comparable across boards; the order is. Spline Arena as of November 18, 2026, AA-3D as of November 17; AA-3D has not rated Gemini 4 Carbon. Sources: Spline Arena rankings; Artificial Analysis AA-3D via Crypto Briefing.2,3,7

The split lands on IPO week

The timing is what makes it money. Anthropic, maker of Claude, is expected to set its IPO price after tonight's close, weighing a price well above $80 against its published $70 to $78 range, and the stock starts trading on Nasdaq as ANT on Thursday, November 19 5.

OpenAI cut Astra's API price 20%, to $8 per million input tokens and $40 output, exactly matching Claude Fable 5.2's list price 6, and has confidentially filed for its own IPO, likely in 2027 5. Alphabet+1.36% — Alphabet, up 1.36 percent today is the listed name already on the board: three of Spline Arena's top four are Gemini 4 variants at $0.98 to $2.01 a scene 2. On Artificial Analysis's broader Intelligence Index, Fable 5.2 still leads Astra 65 to 62 6.

The referee pays the crowd in store credit

Spline's arena doubles as a sales motion. Voters who reach a daily goal claim free Spline AI credits, and their votes may be shared, de-identified, with AI research partners 2. Every battle runs inside the product Spline sells, where agents build through Claude, ChatGPT and Cursor, a coding editor, among clients of MCP, the protocol that lets an AI agent operate an outside tool 1. Spline runs the leaderboard, pays its judges in credits it issues, and keeps the preference data.

Two electorates, two verdicts

The disagreement has a plain explanation. Spline Arena polls open signups judging finished scenes inside Spline's editor; AA-3D polls a panel that Artificial Analysis describes as recruited, judging code in a sandbox 247.

Spline publishes a 95% confidence interval around each rating and treats overlapping ranges as ties 2. Coverage of AA-3D's launch notes that a preference vote reflects the tastes of whoever sits on the panel 4. Read together, the boards say finished-3D quality is not yet one number, and each vendor can quote the board it wins 24.

The first fair test waits on Spline's next model update

The first apples-to-apples test comes when Spline Arena adds GPT-6.2 Astra, whose lineup changes as new models ship 2. Until then the 3D crown has two holders and two scorekeepers, and the price ANT fetches on Thursday prices only one of them 5.

Get Model by email, free

The AI labs, their models, and the business behind them. In our daily email, every weekday.

Loading GOOGL

Reader comments

0 comments

    Model

    HeyGen's Avatar 6 scored 84% on realism, but its 42.2% pick rate was below chance

    Avatar 6 launched November 17, 2026, with an 84% realism figure and first place in a blind arena; the score doubles a below-chance pick rate, and the arena capped 97% of clips at 35 seconds. That is the evidence base under a company claiming $200 million of annual revenue.

    Model

    MiniMax hit a record low, about 90% below its January peak, as a US rule gates its rented Nvidia compute

    Washington's new remote-access rule puts a US license between China's listed model makers and the offshore Nvidia capacity they rent. MiniMax, the purest listed bet on Chinese foundation models, holds the price tag.

    Model

    OpenAI cuts flagship pricing 20% to match Anthropic as heavy users call GPT-6.2 Astra unusable

    OpenAI cut its flagship's price to match Claude Fable 5.2's rate, and the users who run models hardest rejected it, 24 hours before Anthropic priced a $1.4 trillion IPO on the assumption the quality gap holds. The independent labs say the gap is real, for now.

    Model

    Microsoft books Azure revenue on Chinese open models running on Baseten hardware it does not own

    Microsoft now bills customers' committed Azure dollars for open models from Chinese labs running on Baseten's hardware, a toll booth earning rent on compute it does not own.

    Model

    Alibaba stops giving away its best AI model, three days before earnings

    Alibaba launched Qwen4-Max on November 16 and kept the 2.4-trillion-parameter flagship behind its own paid API, promising open weights only for smaller models. Thursday's report is the first test of whether Qwen's three billion downloads become cloud revenue, and it lands two days after a US rule put the chips behind the model under license.

    Model

    Perplexity claims $1 billion of revenue on five times less spend than Anthropic's $1.4 trillion IPO

    Perplexity says it hit $1 billion in recurring revenue on a fraction of Anthropic's spend, three days before Anthropic prices a $1.4 trillion IPO on the opposite assumption. Whose economics hold will only be settled by a filing Srinivas now says could come in early 2027.

    Subscribe to TheInference for free