Three answers on one screen: how Argon trades blows with GPT-6, what to put in your API call (including the 1M output-token continuation), and who can actually call it today. No signup, no scrolling.
Google has not published a model id for the public API, and has not published an opening date. Right now only vetted partners in the Fairwind Program can call it, and resale is not permitted.
The same public benchmarks and independent tests. Only cells with a source.
| Area | Argon | GPT-6 Astra | GPT-6.1 Sol |
|---|---|---|---|
| Writing code: who does it better | |||
| Long-context refactoringFrontierSWE v2 | 55.0% | 65.5% | — |
| Long-horizon engineeringDeepSWE v1.1 | 77.9% | 74.1% | — |
| Finding and fixing bugsCWE-bench v1 | 68.0% | 68.0% | — |
| Business workflow automationZapier AutomationBench | 51.3% | 41.4% | — |
| Long terminal tasksTerminal-Bench 4.0 | 57.4% | — | Opus 5.5 leads at 66.4% |
| Money: what a call really costs | |||
| List price (input / output, per 1M tokens)Argon promo → standard | $2/$10 → $4/$20 |
$10/$50 | $2/$10 |
| Cached inputmatters when you re-feed the same context | $0.10 | $1.00 | $0.10 |
| Output tokens burned per taskArtificial Analysis test, fewer is cheaper | 62,000 | 27,000 | — |
| Real cost per tasksame task on the Intelligence Index, promo pricing | $1.99 | $3.26 | ~$0.74 |
| Reliability and overall | |||
| Composite Intelligence IndexArtificial Analysis, high setting | 53 | 53 | 52 |
| Hallucination rate when wrongAA-Omniscience, lower is better | 15% | 51% | — |
The short version: Argon's input price is 1/5 of Astra's, but it burns 2.3× more output tokens on the same task (62,000 vs 27,000), so real cost per task only falls to about 60% ($1.99 vs $3.26). Astra still wins on long-context refactoring; Argon leads on long-horizon engineering and automation; for the absolute cheapest run, use GPT-6.1 Sol.
The model id is not published yet. The fields below follow the OpenAI-compatible format — on launch day you only change MODEL.
# As of 2026-10-02 there is no gemini-4-argon model id in the public API POST https://generativelanguage.googleapis.com/v1beta/openai/chat/completions -H "Authorization: Bearer $GEMINI_API_KEY" -H "Content-Type: application/json" -d '{ "model": "gemini-4-argon", "messages": [{"role":"user","content":"Refactor this module"}], "max_tokens": 1000000, "temperature": 0.2, "stream": true }'
import os, requests MODEL = "gemini-4-argon" # not published yet, placeholder BASE = "https://generativelanguage.googleapis.com/v1beta/openai" def run(prompt, cont=None): body = {"model": MODEL, "messages": [{"role":"user","content":prompt}], "max_tokens": 1000000} if cont: body["continuation"] = cont # Long Decode Continuation: resume where it stopped r = requests.post(f"{BASE}/chat/completions", json=body, headers={"Authorization": f"Bearer {os.environ['GEMINI_API_KEY']}"}) j = r.json() return j["choices"][0]["message"]["content"], j.get("continuation")