$ skilldirectory
— zsh — registry.skilldirectory.dev
skill libraryblogguidesabout

home / blog / google ai

Gemini 4 Argon vs ChatGPT (GPT-6 Astra and Sol) in 2026: Benchmarks, Pricing and Which to Use

Quick answer: On most public comparisons, Gemini 4 Argon and OpenAI's top models are close. Argon looks cheaper and offers a much longer context in the Gemini app, while OpenAI's best models, GPT-6 Astra and GPT-6.1 Sol, live mostly in ChatGPT Work and Codex.

The independent data we found is mixed and often overlaps, so test both on your own tasks before you pay for either.

"Gemini vs ChatGPT" is hard to answer in late 2026 because each company now sells several models at different prices, and the model you get in the chat window is often not the strongest one. This guide separates the models, shows the numbers with their sources, and flags where the evidence is weak.

Data in this article was gathered on November 18, 2026 from Google's announcements, reporting on OpenAI's releases and third-party benchmark sites. We label who published each number.

The models, in plain terms

ModelMakerWhat it isWhere you can use it
Gemini 4 ArgonGoogleGoogle's frontier model. Announced September 30, 2026Gemini app for Google AI Pro and Ultra subscribers since October 13, 2026; API
GPT-6 AstraOpenAIOpenAI's top model, released September 3, 2026ChatGPT paid plans per OpenAI's rollout, Codex, API
GPT-6.1 SolOpenAICheaper model close to Astra, released September 29, 2026ChatGPT Work and Codex on paid plans; not yet standard chat mode, per one report
GPT-6 SolOpenAIThe default chat model for ChatGPT Plus, per third-party sitesChatGPT

Google says Argon rolled out first to a small group of trusted cyber defenders and then widened. On October 13, 2026, Google said AI Pro and Ultra subscribers could use it in the Gemini app. A Google post on X also lists AI Mode in Search, Antigravity for Ultra subscribers, and the Gemini API for developers and enterprises.

Benchmark charts

Job-level scores (third-party)

BenchLM publishes scores by type of work. It compares Gemini 4 Argon, which it treats as the Google AI Pro chat model, with GPT-6 Sol, the ChatGPT Plus chat model.

Bar chart of BenchLM job scores: Gemini 4 Argon beats GPT-6 Sol on coding, agents, research and reasoning
BenchLM job scores, data verified November 18, 2026. Higher is better.

Argon's point estimates are higher in all four job areas. BenchLM itself calls the overall lead "not settled": Argon scored 79.3 overall and Sol 75.3, and their 90% intervals overlap (74.2 to 84.4 for Argon, 71.0 to 79.6 for Sol). Treat this as "slight edge, not proven."

What the makers report

  • Google (self-reported): Argon scored 77.9% on DeepSWE v1.1, which Google calls state of the art, 51.3% on Zapier's AutomationBench, 91.7% on LVBench for long-video understanding, and tied for first at 68% on CWE-bench v1 for vulnerability remediation.
  • OpenAI (as reported by 9to5Mac): OpenAI described Astra as its most intelligent and aligned model and cited 98% on FrontierMath Tier 4, 99.9% on ARC-AGI-3 and 100% on ExploitBench. We could not verify these numbers independently, and the two companies do not report the same tests, so you cannot compare them directly.

Independent index

Per Artificial Analysis figures cited by Deeplearning.ai's *The Batch*, GPT-6 Astra scored 53 on the Intelligence Index and GPT-6.1 Sol scored 52. Atlas Cloud, citing the same index, put Gemini 4 Argon at 53 and GPT-6 Sol at 48. On those figures, Argon and Astra are essentially tied, and the cheaper GPT-6.1 Sol is within one point.

A small hands-on test

Atlas Cloud ran three prompts on Argon and GPT-6 Sol: a Python duration parser, a policy rewrite, and a flashcard task. The models tied on all three. Sol used far more output tokens because of its reasoning, so the three prompts cost about $0.011 on Argon and $0.12 on Sol. This is a tiny sample, and Atlas Cloud sells access to both models, so use it as a hint rather than proof.

Pricing

Bar chart of API prices per million tokens for Gemini 4 Argon, GPT-6 Sol, GPT-6.1 Sol and GPT-6 Astra
API list prices in US dollars per 1 million tokens.

API (per 1 million tokens, input / output):

  • Gemini 4 Argon: $2 / $10 at the introductory rate, then $4 / $20. BenchLM lists the introductory rate as running through November 30, 2026, and Google's announcement describes the same two price points.
  • GPT-6.1 Sol: $2 / $10, with cached input at $0.10.
  • GPT-6 Sol: $2 / $10.
  • GPT-6 Astra: $10 / $50.

Consumer plans in the US (BenchLM, checked October 16 to November 11, 2026):

TierChatGPTGoogle
EntryGo, $8 a monthAI Plus, $4.99 a month
Main paidPlus, $20 a monthAI Pro, $19.99 a month
TopPro, from $100 a monthAI Ultra, from $99.99 a month

Prices change often, so confirm on each company's plan page before you subscribe.

Context window and limits

Google says it expanded Argon's output limit to 1 million tokens, up from 64,000 before. One price-tracking site lists Argon's context at 1,048,576 tokens. BenchLM says Google AI Pro offers a 1M-token context, while ChatGPT Plus reasoning models list 256K. If you work with very long documents or codebases, that gap matters.

Safety and access notes

Google says Argon is its most resilient model yet against indirect prompt injection and that it released the unrestricted cyber version only to trusted defenders. OpenAI said it limited the first GPT-6 rollout partly because Astra meets its "Critical" cybersecurity threshold, and said it would start with a cybersecurity program. In practice, regular users get the guarded versions.

Who should pick which

  • Pick Gemini (Google AI Pro) if you want a long context window in the chat app, you already live in Google Workspace and Search, or you want the lower price per token on the API.
  • Pick ChatGPT (Plus or Pro) if you use ChatGPT Work or Codex, where OpenAI's strongest models are offered, or you rely on ChatGPT-only tools.
  • Use the API for both if you are building a product. Run your own test set, because the cheapest model that passes your tests is the right one.

A caveat on naming: BenchLM also lists a GPT-6.2 Astra, which we could not confirm from an OpenAI source, so we focused on the models OpenAI's own rollout coverage supports.

FAQ

Is Gemini 4 Argon better than ChatGPT? Not clearly. Third-party scores give Argon a small edge over GPT-6 Sol, but the ranges overlap, and OpenAI's strongest models score about the same as Argon on the one index we found.

Which is cheaper, Gemini 4 Argon or ChatGPT? On the API, Argon's introductory price matches GPT-6 Sol and GPT-6.1 Sol and is far below GPT-6 Astra. The introductory price ends after November 30, 2026, per BenchLM.

Is Gemini 4 Argon free? Google's announcement lists AI Pro and Ultra subscribers, and it does not mention free-tier access.

What is the latest ChatGPT model? OpenAI's top models in our sources are GPT-6 Astra and GPT-6.1 Sol. Check OpenAI's release notes for anything newer.

Which has the longer context window? Gemini. BenchLM lists 1M tokens for Google AI Pro against 256K for ChatGPT Plus reasoning models.

Sources: Google: Gemini 4 Argon, Google: Argon in the Gemini app, BenchLM, The Batch, 9to5Mac on GPT-6 Astra, Atlas Cloud, FutureAGI.