본문으로 건너뛰기

이 리뷰는 아직 한국어 버전으로 제공되지 않아 English 원문을 표시합니다.

Gemini 4 Argon Review: 1M Context, but a 65,536-Token Reply Cap

Pick Gemini 4 Argon for million-token analysis with images and audio, but plan around its 65,536-token reply cap; so far we have run only launch checks.

플레이그라운드 열기

Spicy Index

76.7/ 100

잠정

Freedom Score

아직 테스트 안 함

가격
US$8.00 혼합 100만 토큰당 (입력 3 : 출력 1)
출력 속도
초당 33토큰
성공률
데이터 수집 중

Gemini 4 Argon is Google's new frontier model, announced on 30 September 2026 and aimed at software engineering, enterprise knowledge work and cyber defence. On our API it is one OpenAI-compatible chat endpoint that takes text, images and audio, reads up to 1,048,576 tokens and returns up to 65,536 tokens per reply, with function calling, streaming and three reasoning levels.

The verdict

The most common point of confusion is the output limit. Google's announcement mentions a 1M-token output ceiling, but its developer documentation ties that to a separate long-running mode, while ordinary synchronous calls stop at 65,536 tokens. Our endpoint is the synchronous kind, so plan for 65,536 per reply and split longer work across turns.

This is a launch-day review. On 28 October 2026 our platform ran a real request through the endpoint as part of launch: one neutral message, answered without streaming at low reasoning effort. It returned a complete answer. We also tried video input three ways during integration and it was rejected each time, so this endpoint does not offer video. We have not run a quality blind test or a content-level test, and this page does not score the model.

Try Gemini 4 Argon →

Key facts

  • Tested: one real launch request on 28 Oct 2026 (neutral prompt, low reasoning effort, no streaming), plus integration checks of image and audio input and of video rejection. No scored runs yet.
  • Price: the live price, including cached input, is on the model page.
  • Limits: 1,048,576-token context shared by input and output, replies up to 65,536 tokens (default 4,096, thinking included), text, image and audio input (wav or mp3, 14 MB combined inline), no video or PDF parts.
  • Reasoning: reasoning_effort takes low, medium or high (default medium); thinking cannot be switched off.
  • Sources: scoring rules are on our review methodology page, full parameters on the Gemini 4 Argon model page.

좋았던 점과 아쉬웠던 점

좋았던 점

  • Audio is accepted directly. wav or mp3 can be sent as audiourl or base64 inputaudio next to text and images, so a voice note or call recording sits in the same conversation as your instructions.
  • A million-token window. Contracts, transcripts, a specification and the code it describes can travel together as written.
  • Sampling stays adjustable. temperature (0 to 2) and topp are available, along with stop sequences and responseformat.
  • Tool calling is on the endpoint. Up to 128 function definitions per request with toolchoice and paralleltool_calls; we have not graded tool-use quality yet.
  • Google publishes numbers. Google reports 77.9% on DeepSWE v1.1 (long-horizon software engineering) and 51.3% on AutomationBench. Those are Google's figures; we have not reproduced them.

아쉬웠던 점

  • Replies stop at 65,536 tokens here. The 1M-output figure in Google's announcement does not apply to this endpoint, and thinking counts toward the 65,536.
  • No video input. Google lists video among the model's input types, but our three integration attempts were all rejected, so frames or a separate audio track are the workaround.
  • No PDF parts. Send pages as images or extracted text.
  • Thinking cannot be turned off, and low is the lightest setting, so short lookups still pay for a reasoning pass.
  • No quality numbers of our own yet. We have not run a blind test, a speed test or a long-document test.

What we tested

On 28 October 2026 our platform sent Gemini 4 Argon one real request as part of launch: a single neutral user message, answered without streaming at low reasoning effort with a 1,024-token cap. It returned a complete answer, and the request was priced at the published rate. The endpoint also passed our pre-launch checks of its price table and request limits.

During integration, image and audio input were accepted and answered. Video was different: a one-second clip with sound, a five-second silent clip and the same clip sent inline each came back as a rejected request, so the endpoint declares only images and audio.

That is a working-endpoint check, not a benchmark. We have not compared Argon's answers with another model, measured speed, or tried a prompt near the 1,048,576-token limit.

Prompts to start from

These starting prompts come from the model page's published examples. We have not graded the outputs, so they are a place to begin rather than prompts proven on this model.

1. Compare a transcript with a brief. Paste both documents, label them, and say which is authoritative.

text
Compare a meeting transcript with the product brief and list unresolved decisions. Work from the material I provide, cite the relevant evidence, distinguish facts from assumptions, and identify missing information before proposing changes.

Run this prompt on Gemini 4 Argon →

2. A tool step. Confirms your function-calling loop end to end with a trivial function.

text
Call report_status with status ready.

Settings that matter: raise maxtokens as you raise reasoningeffort, because thinking is paid for out of the same 65,536; send audio inline and keep it under 14 MB combined; for work longer than one reply, ask for it in parts.

How far it writes

We have not run our content-level test on Gemini 4 Argon, so we have no result to report and the endpoint carries no Unrestricted label. The model page states the position plainly: content decisions come from the model, behaviour can vary, and no guarantee of unrestricted output is made.

Not tested yet: the content-level ladder, multi-turn behaviour over a long session, languages other than English, audio quality on noisy recordings, and prompts close to the context limit.

Every character in a request should be an adult, keep real, identifiable people out of it, and you are responsible for the law where you publish.

What the community says

Reaction on the launch thread was cautious about the benchmarks. One reader wrote "Big number results, and impressive pricing. That said it really feels like benchmarks have been hyper saturated these days." Another was blunter about where they would use it: "Hate to say i will never be touching this model for anything except for youtube video understanding", which is the one input type our endpoint does not offer.

Alternatives

  • Hard coding with a larger reply ceiling? GPT 6.1 Sol allows replies up to 128,000 tokens and takes images but not audio.
  • A lighter Google model? Gemini 3.7 Flash is the Flash-class option from the same publisher.
  • An Anthropic flagship? Claude Opus 5.5 is the Anthropic option on the same API; compare it on your own prompts.

Gemini 4 Argon의 Spicy Index 구성

아래 각 항목에서 채점 세부 내용으로 이동할 수 있습니다. 모든 점수가 어느 행에서 나왔는지, 왜 감점됐는지 확인하세요.

공개 리더보드 품질의 40%
아직 테스트 안 함

추적 중인 공개 리더보드에 아직 정확히 일치하는 항목이 없습니다. 작동 방식

롤플레이 품질의 15%
아직 테스트 안 함

이 모델에는 아직 롤플레이 테스트를 돌리지 않았습니다. 작동 방식

창작 글쓰기 품질의 15%
61.6
지시 이행 품질의 15%
99.0
긴 컨텍스트 기억 품질의 15%
아직 테스트 안 함

이 모델에는 아직 긴 컨텍스트 테스트를 돌리지 않았습니다. 작동 방식

기능 가중치 10.5%
56.0
Freedom Score 가중치 30%
아직 테스트 안 함

자유도가 미측정이거나 근거가 부족합니다. Index는 원래 품질·기능 점수를 유지하고 잠정 표시로 순위에 포함됩니다. 근거가 없다고 0점으로 처리하지 않습니다. 작동 방식

자유도가 미측정이거나 근거가 부족합니다. Index는 원래 품질·기능 점수를 유지하고 잠정 표시로 순위에 포함됩니다. 근거가 없다고 0점으로 처리하지 않습니다.

전체 테스트 리포트 보기 →

기능 항목별 점수

공개 채점표의 점수입니다. 업스케일로 만든 해상도나 당사가 자체적으로 둔 제한은 반영하지 않습니다.

컨텍스트 창(개발사 공개값)명시 안 됨
0.0 / 30
SpicyAPI 입력 형식텍스트와 이미지, 오디오
19.0 / 25
도구 사용도구와 tool_choice
15.0 / 15
구조화된 출력지원, 스키마 준수 확인
15.0 / 15
추론 제어3단계: low · medium · high
7.0 / 15

SpicyAPI에서 지원하는 Gemini 4 Argon 기능

SpicyAPI의 실시간 모델 사양에서 그대로 가져왔습니다. 개발사가 자사 앱에서 더 많은 기능을 제공할 수도 있지만, 여기 있는 것은 지금 당사 API로 호출할 수 있는 기능입니다.

엔드포인트별 기능
기능채팅
컨텍스트 창1,048,576토큰
요청당 최대 출력65,536토큰
추론 단계low · medium · high
도구 호출예
구조화된 출력예
모든 파라미터와 코드 예시 보기 →

실제로 걸리는 시간

지난 30일 동안의 실제 작업에서 요청부터 완성까지를 측정했습니다. 시간에는 당사가 직접 돌린 테스트 작업도 포함합니다.

출력 속도
초당 33토큰첫 토큰까지의 시간 포함
성공률
데이터 수집 중

완료된 작업이 10건 이상 모이면 시간을 표시합니다. 성공률은 고객 작업만으로 계산하며, 30일 분량의 트래픽이 모일 때까지 모든 모델에서 숨겨 둡니다.

Gemini 4 Argon에 대한 사람들의 이야기

긍정 0 · 엇갈림 1 · 부정 1

  • Big number results, and impressive pricing. That said it really feels like benchmarks have been hyper saturated these days.

    hacker-news· 2026년 9월 30일· 엇갈림출처

  • Hate to say i will never be touching this model for anything except for youtube video understanding

    hacker-news· 2026년 9월 30일· 부정출처

인용문은 공개 게시물을 그대로 옮기고 원문에 링크했습니다. 평가가 아니며 어떤 점수에도 반영하지 않습니다.

Gemini 4 Argon 대신 고려할 만한 모델

어떻게 생각하세요?

이 리뷰가 도움이 되었나요?

투표하려면 로그인하세요.

댓글

댓글은 자동 검사와 당사 팀의 확인을 거칩니다. 스팸, 부적절한 내용, 인신공격, 동의 없이 실존 인물을 거론하는 글은 삭제합니다. 댓글을 신고하면 영업일 기준 하루 안에 검토합니다.

이 모델을 써 보셨나요? 잘 된 방법을 공유해 주세요. 스팸을 막기 위해 댓글을 달려면 무료 계정이 필요합니다.

Gemini 4 Argon 자주 묻는 질문

Gemini 4 Argon 직접 써 보기

플레이그라운드에서 모든 파라미터를 설정하고 실행 전에 정확한 가격을 확인하세요.

US$8.00 혼합 100만 토큰당 (입력 3 : 출력 1)

다른 리뷰

리더보드 보기 →
  • Small blank marble tiles race along a short brass rail on a basalt plinth into a three-slot marble gate, where a gold filament splits into three channels ending in three neat stacks

    GPT 6.1 Terra

    Pick GPT 6.1 Terra for high-volume screens, documents and writing where reasoning_effort none helps; so far we have run only launch checks, no scored tests.

    Spicy Index
    79.9
    Freedom Score
    —

    2026년 10월 28일 테스트

    리뷰 읽기
  • Small blank marble tiles race along a short brass rail on a basalt plinth into a three-slot marble gate, where a gold filament splits into three channels ending in three neat stacks

    Kimi K3.1 Pro

    Pick Kimi K3.1 Pro for long-context work with screenshots and clips when you accept always-on thinking; so far we have run only launch checks, no scored tests.

    Spicy Index
    80.4
    Freedom Score
    —

    2026년 10월 28일 테스트

    리뷰 읽기
  • A burst of golden light filaments fanning low across a shallow glass dish against pure black

    Grok 4.3

    거대한 문맥 창, 진짜 추론 끄기 스위치, 저렴한 무검열 글쓰기가 필요하면 Grok 4.3을 고르세요. 파일 업로드가 필요하면 건너뛰세요.

    Spicy Index
    77.0
    Freedom Score
    98.9

    2026년 9월 24일 테스트

    리뷰 읽기