Last updated:
Gemini 4 Argon vs GPT-6 Astra
Google's new frontier model against OpenAI's flagship. Both are the top model of their maker; Argon costs far less per token.
Side by side
| Gemini 4 Argon | GPT-6 Astra | |
|---|---|---|
| Vendor | Google DeepMind | OpenAI |
| Released | September 30, 2026 (Fairwind Program only) | September 4, 2026 |
| Input / output per 1M | $2 / $10 launch, $4 / $20 standard | $10 / $50 |
| Cached input per 1M | $0.10 launch, $0.20 standard | $1 |
| Max output | 1M tokens | 128K tokens |
| Context window | — | 1.05M |
Benchmarks
From Google's Gemini 4 Argon table. Google ran every model at its highest thinking setting. OpenAI publishes its own comparisons, which can differ.
| Benchmark | Gemini 4 Argon | GPT-6 Astra |
|---|---|---|
| Vals Index | 68.9% | 63.1% |
| AutomationBench | 51.3% | 41.4% |
| Vals Finance Agent v2 | 65.4% | 53.5% |
| Harvey's Legal Agent Benchmark | 19.6% | 5.4% |
| DeepSWE v1.1 | 77.9% | 74.1% |
| FrontierSWE v2 | 55.0% | 65.5% |
| Vibe Code Bench | 91.9% | 89.6% |
| Terminal-Bench 4.0 | 57.4% | 58.2% |
| PostTrainBench | 45.3% | 44.3% |
| Terminal-Bench Science 0.1 | 57.6% | 68.1% |
| LABBench 2 | 88.8% | 85.4% |
| RiemannBench | 76.0% | 72.0% |
| GraphWalks BFS (up to 128K) | 99.7% | 98.7% |
| GraphWalks BFS (256K-1M) | 84.2% | 71.8% |
| Agent's Last Exam | 39.5% | 34.2% |
| OSWorld-2.0 | 69.2% | 72.6% |
| Chartography | 71.6% | 71.0% |
| LVBench | 91.7% | 87.5% |
| CWE-bench v1 | 68.0% | 68.0% |
Argon leads on 14, GPT-6 Astra on 4, and 1 is a tie.
Source: Google DeepMind: Gemini model page
Cost per coding task
| Codebase size | Argon (launch) | Argon (standard) | GPT-6 Astra |
|---|---|---|---|
| Small (< 10k lines) | $0.112 | $0.223 | $0.576 |
| Medium (10k-100k lines) | $0.249 | $0.498 | $1.290 |
| Large / monorepo (> 100k lines) | $0.594 | $1.188 | $3.090 |
At 173 tasks a month on a medium codebase: about $43 on Argon at the launch price ($86 at the standard price) and $223 on GPT-6 Astra, with the same token counts for both.
Gemini 4 Argon: full details Compare every model in the calculator
Which one to pick
- Cost: Argon, at $4 / $20 standard against $10 / $50.
- Hard software-engineering tasks: Astra led on FrontierSWE v2 (65.5% vs 55.0%); Argon led on DeepSWE (77.9% vs 74.1%). Test both on your own repo.
- Business automation and long context: Argon, on AutomationBench (51.3% vs 41.4%) and GraphWalks 256K-1M (84.2% vs 71.8%).
- Need it today: Astra. Argon is limited to Google's Fairwind security partners for now.
FAQ
Is Gemini 4 Argon better than GPT-6?
On Google's table, Argon beats GPT-6 Astra on 14 of 19 shared benchmarks and ties on 1. Astra is ahead on FrontierSWE v2, Terminal-Bench 4.0, Terminal-Bench Science 0.1, OSWorld-2.0.
Which is cheaper, Gemini 4 Argon or GPT-6 Astra?
Argon: $2 / $10 at launch and $4 / $20 after, against $10 / $50 for Astra.
Gemini 4 Argon vs GPT-6.1 Sol?
GPT-6.1 Sol has the same $2 / $10 price as Argon's launch price and is available now. Google's table does not include it.
Independent page, not affiliated with Google, OpenAI or Meta. Benchmarks are Google-published; we have not run our own tests.