Gemini 4 Neon Tops LiveBench at 87.4, 2.5 Points Ahead

Google’s Gemini 4 Neon is first on the LiveBench 2.0 leaderboard with an overall score of 87.4, ahead of Anthropic’s Claude Fable 5.2 and OpenAI’s GPT-6.2 Astra, which are tied at 84.9.

That is a 2.5-point lead, by our arithmetic. LiveBench also lists the cost per successful task: $2.654 for Neon against $9.850 for Claude Fable 5.2 and $1.941 for GPT-6.2 Astra.

It is one leaderboard among several, and they do not agree. Our AI model benchmarks page tracks how the top models compare across the main tests.

Business Pill · The Model and the Harness

An evergreen Business Engineer explainer. It teaches the general idea only and does not discuss this leaderboard.

The key insight: LiveBench 2.0’s leaderboard, read on 22 November, puts Google’s Gemini 4 Neon first at 87.4, with Claude Fable 5.2 and GPT-6.2 Astra tied at 84.9; Neon costs $2.654 per successful task against $9.850 for Fable 5.2.

What LiveBench Measures

LiveBench, run by Abacus AI, gives models real work and grades what they hand back, it says. Its board has seven suites, according to its methodology page. Five are agentic: the model uses tools to finish end-to-end tasks, such as fixing a real bug in a large codebase or completing jobs in web apps through a browser.

The other two, Data Analysis and Reasoning, are short sets of questions with one correct answer. Each score is the model’s average from a single run, from 0 to 100. The overall score counts the long agentic suite twice and coding and design one and a half times.

Gemini 4 Neon leads LiveBench 2.0 at 87.4; Claude Fable 5.2 and GPT-6.2 Astra are tied at 84.9, 2.5 points beh
Gemini 4 Neon leads LiveBench 2.0 at 87.4; Claude Fable 5.2 and GPT-6.2 Astra are tied at 84.9, 2.5 points behind by our arithmetic.

Where Each Model Wins

Neon leads Browser Use with 95.6 and ties Claude Fable 5.2 on Research at 94.5. GPT-6.2 Astra has the best Coding score of the top three, 91.9. Claude Fable 5.2 has the best Agentic score of the three, 82.9, and the best Design score, 85.1.

Astra’s weak spot is the long agentic suite, where it scores 70.8, about 12 points below Neon and Fable 5.2 by our arithmetic. That suite carries double weight in the overall score.

The Cost Gap

On LiveBench’s cost measure, Claude Fable 5.2 costs about 3.7 times as much as Neon per successful task, by our arithmetic. GPT-6.2 Astra is the cheapest of the three at $1.941, about 27% below Neon.

What Other Leaderboards Say

The order changes from board to board. On Arena’s text leaderboard, Google holds the top three places, but Gemini 4 Carbon leads and Neon is third, as our fact-check of a viral OpenAI post noted. On the Artificial Analysis index, Claude Fable 5.2 scored 65 and GPT-6.2 Astra 62.

On Arena’s Agent Arena Index, Neon ranked second with 1,336 points. And on Arena’s Alignment Index, OpenAI models hold the top seven places, led by GPT-6.2 Sol at 85.2.

The Structural Read

As we read it, the long agentic suite, which counts double, is what separates the top three.

As we read it, the spread between leaderboards says more about what each one tests than about a single best model.

LiveBench

“A challenging, contamination-free long-horizon AI benchmark. Models do real work (decks, research reports, bug fixes, apps that run for hours) and we grade what they hand back.”

Three Implications

FOR TEAMS CHOOSING MODELS LiveBench lists cost per successful task alongside scores: $1.941 for GPT-6.2 Astra, $2.654 for Neon and $9.850 for Claude Fable 5.2.

FOR GOOGLE Gemini 4 Neon leads LiveBench overall and on Browser Use, while Gemini 4 Carbon leads Arena’s text board.

FOR OPENAI AND ANTHROPIC GPT-6.2 Astra has the best Coding score of the top three; Claude Fable 5.2 has the best Agentic and Design scores.

What Is Not Established

LiveBench scores come from a single run per suite, and the page says it was last updated on 11 November. A task the run could not measure scores zero, it says. Rankings on any one board can move when a new model or a new release is added.

This Week In The AI Supercycle

A World Without an Order, and an AI Central Bank Without a Successor

This week’s issue: the week AI memory stayed largely sold out while the people holding the shares paid: why the AI Supercycle is not a dot-com replay, read through geopolitics, macro and technology, with the gap between building and financing the widest we’ve recorded.

Read the full issue →

Business Engineer Framework

The AI Supercycle

The Business Engineer map of where AI capability turns into business leverage.

Read the Map of AI →

The Bottom Line

On LiveBench’s mix of real work tasks, Gemini 4 Neon leads by 2.5 points and costs far less per success than Claude Fable 5.2. Other boards rank the same models differently, so no single leaderboard settles which model is best.

97,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.

A note on sourcing. Scores, suites, weights and cost per successful task are from LiveBench’s leaderboard and methodology pages, read on 22 November; the page says it was last updated on 11 November. The 2.5-point gap, the 3.7 times cost ratio and the 27% figure are our arithmetic. The Arena and Artificial Analysis figures are as reported in our earlier pieces, linked above.

Sources: LiveBench 2.0 leaderboard, release 2026-11-06, read 22 Nov 2026 · LiveBench methodology page · FourWeekMBA fact-check of a viral OpenAI post, 21 Nov 2026 · FourWeekMBA on Gemini 4 Neon’s Agent Arena rank, 20 Nov 2026 · FourWeekMBA on Arena’s Alignment Index, 22 Nov 2026

Scroll to Top
FourWeekMBA