Gemini 4 Argon parameters and benchmarks: an evidence-led comparison
Gemini 4 Argon combines a 1,048,576-token unified context window with multimodal input, long decoding and tool use. Its official table is strong but not a sweep.
Parameters that matter
Google documents a 1,048,576-token unified input-and-output limit, 65,536 full-speed synchronous output tokens, and up to 1,048,576 output tokens through long decoding. Inputs include text, images, video, audio and PDF; output is text.
Architecture and tools
The model card describes a sparse mixture-of-experts transformer with native text, vision and audio support. The API documents function calling, file search, code execution, search grounding, structured output and computer use in preview.
Where Argon leads
In Google’s table, Argon scores 68.9% on Vals Index, 65.4% on Vals Finance Agent v2, 19.6% on Harvey’s Legal Agent Benchmark, 77.9% on DeepSWE v1.1 and 91.7% on LVBench.
| Benchmark | Gemini 4 Argon | GPT-6 Astra | Claude Fable 5.1 | Claude Opus 5.5 |
|---|---|---|---|---|
| Vals Index | 68.9% | GPT-6 Astra 63.1% | Claude Fable 5.1 65.8% | Claude Opus 5.5 67.0% |
| DeepSWE v1.1 | 77.9% | GPT-6 Astra 74.1% | Claude Fable 5.1 67.4% | Claude Opus 5.5 74.2% |
| FrontierSWE v2 | 55.0% | GPT-6 Astra 65.5% | Claude Fable 5.1 56.3% | Claude Opus 5.5 62.3% |
| Terminal-bench 4.0 | 57.4% | GPT-6 Astra 58.2% | Claude Fable 5.1 57.9% | Claude Opus 5.5 66.4% |
| GraphWalks 256k–1M | 84.2% | GPT-6 Astra 71.8% | Claude Fable 5.1 65.0% | Claude Opus 5.5 66.8% |
| LVBench | 91.7% | GPT-6 Astra 87.5% | Claude Fable 5.1 79.7% | Claude Opus 5.5 83.7% |
Where competitors lead
Argon is not first everywhere. GPT-6 Astra leads FrontierSWE v2 and Terminal-Bench Science 0.1; Claude Opus 5.5 leads Terminal-bench 4.0 and PostTrainBench; GPT-6 Astra also leads the OSWorld-2.0 row.
How to evaluate it yourself
Benchmark settings differ from real workloads. Test the same prompt set, files, tool permissions and review rubric across models, record failures and cost, and do not compare a vendor score with a differently configured independent harness as if they were identical.
Start with a real task
This workspace obtains Argon’s capability flags and quote from this site's live catalog at runtime. Select the model, submit a representative task, then inspect the result and ledger rather than extrapolating from a single benchmark.
Run your own comparison
Use the same prompt and evidence checklist, then compare the result and final ledger entry.
Test Gemini 4 Argon