Tabbit
ResourcesBlogModels
Tabbit LogoTabbit

Tabbit — The AI Browser that Works for You

Topics

  • Best Browsers 2026: Rankings and Top Lists
  • AI Browser Guides and Resources
  • Browser Comparisons: Head-to-Head Guides
  • Browser Reviews and Explainers
  • Browser Alternatives
  • Browser Directory: Guides by Browser
  • Agentic Browser Resources
  • Browser Downloads and Install Guides
  • Browser Productivity Resources
  • SillyTavern Model Guides and Presets

Popular Guides

  • Best Browsers 2026: Top 10 Ranked & Compared
  • Best Free AI Browsers 2026: Plans & Prices Compared
  • Rabbit Browser: What It Is & Why Tabbit Is Better
  • Firefox vs Chrome 2026: Privacy, RAM & Speed Compared
  • Best Lightweight Browsers 2026: 7 Tested for RAM & Speed
  • AI Browser List 2026: 10 Browsers Compared
  • Neobrowser: Norton Neo, Download & Alternatives
  • Edge vs Chrome 2026: Speed, RAM & Battery Tested
  • Best Internet Browser for Android 2026
  • Best AI Browser 2026: Top 9 Tested & Ranked

Events

  • Tabbit Skill Competition
  • KPOP SBTI Fandom Personality Test
  • Tabbit Campus Creator Program
  • fifi's Picks: AI Skills for Research Papers
  • User Survey

About

  • Tabbit Blog
  • Press & Media
English
简体中文English
Review
MediaGemini 4 Carbon

Gemini 4 Carbon's Composite Score and Benchmark Coverage on AI IQ

Original source

AI IQ

AuthorNot specified (the page is attributed to AI IQ)

Tabbit curation2026-11-16

Read original

One-sentence takeaway

As of 2026-11-16, AI IQ gives Gemini 4 Carbon a Composite IQ of 142, ranking it #2/180. Results are published for 13 of its 39 benchmarks, covering 6/7 dimensions. The 142 is an AI IQ metric produced by calibration and estimation, with equal weighting across seven dimensions. It is not a human IQ measured directly from the model.

Use cases

  • Tasks it can help assess: Reviewing the composite score, dimension scores, rankings, and published benchmark coverage according to AI IQ's own methodology, and comparing models within the same snapshot on that site.

  • Claims it cannot support: It does not establish that the model has a human IQ of 142 or support inferences about its ability on untested tasks. The AI IQ page should not be treated as the original publisher of the 13 benchmarks or as an independent retest record.

  • Model version covered: Gemini 4 Carbon as labeled on the AI IQ page. The page does not identify the specific model revision or reasoning tier used for each benchmark.

  • Test environment or client: AI IQ's public model profile and the public benchmark leaderboards it references. The model page does not provide a complete harness, sample size, run parameters, or raw outputs for each benchmark.

  • Reasoning tier and parameters: Not specified.

Evaluation method

AI IQ maps 39 benchmarks to seven dimensions: abstract reasoning, mathematical reasoning, academic reasoning, programming reasoning, computer use, reliability, and knowledge work. The methodology page says each raw benchmark score is converted using piecewise linear interpolation against preset expected scores at IQ 70, 85, 100, 115, 130, 145, and 160 for that benchmark. The six dimensions other than abstract reasoning use a Bayesian model that combines observed results with uncertain priors from predecessor models. Abstract reasoning is estimated from direct evidence. The seven dimensions are then averaged with equal weights:

[ \mathrm{Composite\ IQ}=\frac{1}{7}\sum_{d=1}^{7}\mathrm{IQ}_d ]

On the model page, “6/7 dims” means benchmark measurements are available as inputs for six dimensions. “13/39 benchmarks” means results have been published for 13 of the 39 benchmarks. Academic reasoning is an estimate, and there are currently no published benchmark results for that dimension. The methodology page says all seven dimension scores are estimates, even when benchmark coverage is complete. Missing values enter only the derived scoring process and are not presented as observed benchmark scores. For some missing entries, predecessor information such as Gemini 4 Argon may be used as a reference.

Key results

DimensionAI IQ scoreStatusBenchmark coverageDimension ranking
Abstract reasoning134Benchmark inputs available, score is a derived estimate3/5#3/209
Mathematical reasoning147Benchmark inputs available, score is a derived estimate2/5#5/180
Academic reasoning136Estimated0/6#3/211
Programming reasoning150Benchmark inputs available, score is a derived estimate1/6#6/181
Computer use154Benchmark inputs available, score is a derived estimate1/5#2/180
Reliability139Benchmark inputs available, score is a derived estimate3/6#2/181
Knowledge work134Benchmark inputs available, score is a derived estimate3/6#2/180

The page says reliability and computer use have the highest dimension rankings (both rank second), while programming reasoning ranks relatively lower (sixth). The number of models differs by dimension, so rankings should be interpreted within each dimension.

Raw data

The AI IQ model page lists the following 13 results. Scores and individual rankings are the values currently shown on the aggregation page. The upstream source column identifies the data source listed on the page. It does not mean that AI IQ retested these benchmarks under a unified setup.

DimensionBenchmarkScore listed on pagePage rankingSource listed on page
Abstract reasoningARC-AGI-294.2%#2/71Full source leaderboard checked, Nov 13
Abstract reasoningARC-AGI-198.5%#5/71Full source leaderboard checked, Nov 13
Abstract reasoningARC-AGI-365.1%#2/26Full source leaderboard checked, Nov 13
Mathematical reasoningFrontierMath Tier 495.1%#4/75Full source leaderboard checked, Nov 13
Mathematical reasoningFrontierMath Tier 1–390.5%#6/87Full source leaderboard checked, Nov 13
Programming reasoningTerminal-Bench 4.065.2%#2/109Artificial Analysis v5.0.1, Nov 13
Computer useArena.ai Agent Arena17.0#3/72Arena.ai Agent Arena, Nov 13
ReliabilitySimpleQA Verified77.7%#2/83Full source leaderboard checked, Nov 13
ReliabilityBullshitBench v272.0%#10/88Full source leaderboard checked, Nov 13
ReliabilityFACTS Grounding77.2%#6/45Full source leaderboard checked, Nov 13
Knowledge workAutomationBench-AA77.8%#1/112Artificial Analysis v5.0.1, Nov 13
Knowledge workGDP.pdf28.4%#7/111Artificial Analysis v5.0.1, Nov 13
Knowledge workAA-Briefcase v1.1 Elo1,611#10/116Artificial Analysis v5.0.1, Nov 13

The remaining 26 benchmarks without listed scores are marked “Not yet reported” on the AI IQ page, not scored as zero. None of the six academic reasoning benchmarks has been reported. The current page also marks effective cost, speed, and context as unverified or unmeasured. These gaps are not evidence for or against the composite score conclusion in this note.

Conclusions and limitations

  1. 142 is a derived score from AI IQ. The seven dimensions are equally weighted, but published benchmark coverage is uneven. Six dimensions have result inputs, while the academic reasoning score is estimated. Many dimensions also have results for only some of their benchmarks.

  2. Benchmark scores and Composite IQ are different layers of data. The 13 results in the table are upstream benchmark results collected by AI IQ. The 142 and dimension scores are generated through its calibration scale, estimation, and aggregation rules. AI IQ is the direct source for its own summary scores and estimates, but it is not the original publisher of the underlying benchmarks listed here.

  3. The methodology page explicitly acknowledges calibration limitations. The mapping from benchmarks to IQ includes subjective design choices. The page says intervals are estimates under model assumptions, not empirically calibrated confidence intervals. It also explicitly describes IQ as an analogy for ease of understanding, not as equivalent to a standardized human IQ test.

  4. Rankings are a changing snapshot. Changes to the leaderboard, upstream benchmarks, or calibration rules can alter scores or rankings. This note records only what was visible on 2026-11-16 and does not generalize it into a conclusion about model intelligence across tasks.

Reproduction notes

Open the Gemini 4 Carbon model page on AI IQ and record the access date, Composite IQ, dimension table, and coverage across all 39 benchmarks. Also open AI IQ Methodology to check the seven-dimension framework, interpolation, and missing-value explanations. To verify the underlying scores, consult the upstream platforms listed on the page and the original leaderboards for the Nov 13 data snapshot, preserving their versions, scoring methods, and harness differences. The two AI IQ pages do not provide the inputs, run parameters, and raw outputs needed to reproduce the 13 evaluations.

Curated by Tabbit

This is a third-party source navigator. Model versions, test environments, and personal experience vary; consult the original source.

Gemini 4 Carbon

Use and compare models in Tabbit

Gemini 4 Carbon

Related reviews

OfficialGoogle DeepMind

Gemini 4 Carbon Model Card: Knowledge Work and Multimodal Benchmarks and Their Limits

MediaArtificial Analysis

Gemini 4 Carbon: Artificial Analysis Reasoning Tiers and Index Costs

MediaZapier

Gemini 4 Carbon on Zapier AutomationBench: Business Workflow Results

MediaCursor's official CursorBench leaderboard2026-09-10

CursorBench 4.0: Gemini 4 Carbon on Multi-File Engineering Tasks

Gemini 4 Carbon

Related showcases & prompts

OfficialGoogle AI Studio / Gemini API

Google AI API Specifications for Gemini 4 Carbon

OfficialGoogle AI Studio / Gemini API

Gemini 4 Carbon Long Outputs: Continuation and Stop Conditions

OfficialGoogle AI Studio2026-11-10

Google AI Studio: Prompting Gemini 4 Carbon for Complex Tasks

OfficialGoogle AI Studio / Gemini API

Google AI Studio: Gemini 4 Carbon Agentic Video Understanding Configuration