AI IQ gives Gemini 4 Hydrogen a Composite IQ of 131, ranking it 16th among 180 models. This is not a score from a single test: it is an equal-weighted aggregate of seven dimensions, five of which are estimated. The four publicly listed benchmark results are the scores directly attributable to tests.
Tasks it can help assess: Reviewing coding and knowledge-work benchmark results, estimated dimensions, and data coverage gaps recorded by AI IQ.
Tasks it cannot support extrapolating to: Treating Composite IQ as a general capability score, treating estimates as measured results, or using them to guarantee performance on specific agents or production tasks.
Model version covered: Gemini 4 Hydrogen; the AI IQ page does not tie the benchmark results to a specific reasoning tier.
Test environment or client: The benchmark source is listed as Artificial Analysis v5.0.1 (2026-11-13); the specific runtime environment, harness, and model endpoint are not stated.
Reasoning tier and parameters: Not stated.
AI IQ takes the equally weighted average of seven dimensions—abstract reasoning, mathematical reasoning, academic reasoning, coding reasoning, computer use, reliability, and knowledge work—to calculate Composite IQ. The page defines 100 as average human performance. Model dimension scores and rankings are calculated against the models scored in each dimension.
The page records public Gemini 4 Hydrogen results for 4 of the 39 Composite IQ benchmarks; 35 have no results yet. Coding reasoning and knowledge work are marked as measured based on published results. The other five dimensions are estimated from the model's results in other dimensions and information about earlier models such as Gemini 3.8 Flash. AI IQ specifically notes that earlier-model scores are inputs to estimates and must not be treated as measured results for Hydrogen. The displayed 95% ranges are uncalibrated model-estimate ranges, not statistical confidence intervals.
| Item | AI IQ score | Rank | Evidence status |
|---|---|---|---|
| Composite IQ | 131 | #16 / 180 | Equal-weighted across 7 dimensions; includes estimates |
| Abstract reasoning | 126 | #10 / 209 | Estimated, 0/5 results available |
| Mathematical reasoning | 127 | #42 / 180 | Estimated, 0/5 results available |
| Academic reasoning | 132 | #20 / 211 | Estimated, 0/6 results available |
| Coding reasoning | 143 | #14 / 181 | Measured dimension, 1/6 results available |
| Computer use | 124 | #35 / 180 | Estimated, 0/5 results available |
| Reliability | 130 | #11 / 181 | Estimated, 0/6 results available |
| Knowledge work | 133 | #7 / 180 | Measured dimension, 3/6 results available |
| Benchmark | Score | Rank | Source listed on the page |
|---|---|---|---|
| Terminal-Bench 4.0 | 53.5% | #11 / 109 | Artificial Analysis v5.0.1, Nov 13 |
| AutomationBench-AA | 76.1% | #4 / 112 | Artificial Analysis v5.0.1, Nov 13 |
| GDP.pdf | 27.6% | #8 / 111 | Artificial Analysis v5.0.1, Nov 13 |
| AA-Briefcase v1.1 Elo | 1,567 | #17 / 116 | Artificial Analysis v5.0.1, Nov 13 |
These benchmark results and AI IQ scores are on different scales. AI IQ says the benchmark scores in the table retain the values from their source data. The page does not provide the original prompts, complete run logs, scoring details, or parameters for each test, so the tests cannot be independently recalculated here, nor can the significance of differences across models be confirmed.
The page summary gives Composite IQ 131, #16 / 180, and says the model outperforms 92% of ranked models.
The page summary bar says 2/7 dimensions are measured and 4/39 benchmarks have public results; the other five dimensions are estimated.
The page says Hydrogen is strongest in knowledge work and coding reasoning, at 133 (#7 / 180) and 143 (#14 / 181), respectively. These dimension values are still scores on AI IQ's scale, not raw AutomationBench or Terminal-Bench percentages.
The listed model prices are $1.50 per million input tokens and $7.50 per million output tokens. Effective cost, calculated from token consumption on real tasks, is $15.92 per million tokens. Effective cost is an aggregate metric from the site; the page does not provide its calculation details.
AI IQ marks speed as unmeasured. The context information is internally inconsistent: the top bar says “Unverified / Not in publisher docs,” while the release information section says the shared input and output context budget is 1,000,000 tokens. This conflict cannot be resolved from the page alone.
AI IQ's data is useful as a dated third-party benchmark overview. Of the four public results, knowledge work has coverage for 3/6 benchmarks and coding reasoning for 1/6; the other five dimensions have no public results for Hydrogen. Composite IQ 131 therefore combines measured inputs with model estimates and should not be read as an aggregate measured result across all 39 benchmarks.
The ranking denominator varies by dimension (180 to 211 models). For comparisons, use the same dimension and the benchmarks listed on the page rather than relying only on rank or the composite score. Benchmark versions, harnesses, and task composition can affect results; AI IQ also cautions that small differences between models may not be meaningful when source evaluations use different settings. The page does not disclose the reasoning tier used for Hydrogen's tests, so the scores cannot be attributed to a particular tier.
The original AI IQ model page can be checked against the collection date for its summary score, dimension coverage, four benchmark results, source labels, and estimation notes. To rerun the raw benchmarks, first obtain the same benchmark versions and harnesses, then fix the model endpoint and reasoning settings. The AI IQ page does not provide the prompts, run traces, and complete parameters needed to reproduce the original experiments. Report rerun results separately from the AI IQ estimated scores in this note.
Gemini 4 Hydrogen