Gemini 4 Argon vs Gemini 4 Carbon
Everything below is quoted from Google’s official announcements and documentation (Argon launch post, Carbon launch post, Argon model page), checked on 2026-10-29. Prices are Google’s list prices, not byesu’s; on byesu you are billed per token and see the actual price in the model list after you sign in.
Short answer
If you are building against the API today, Gemini 4 Argon is the model you can actually call — it has been generally available since 2026-10-13 under the stable ID gemini-4-argon. Gemini 4 Carbon, announced 2026-10-28, is Google’s newer and stronger model, but it is in a phased rollout that starts with cyber defenders and trusted partners; Google has not published a model ID, a developer API date or a Carbon model page. Both launch at the same introductory price.
At a Glance
| Item | Gemini 4 Argon | Gemini 4 Carbon |
|---|---|---|
| Announced | 2026-10-13 | 2026-10-28 |
| Google’s positioning | Flagship reasoning model “built to solve your hardest problems”: complex software engineering, deep math and science, autonomous multi-step agents, long decode up to 1M tokens | “Our strongest model with state-of-the-art cybersecurity and long horizon capabilities” |
| Developer API availability | Generally available (AI Studio, Gemini API, Gemini Enterprise Agent Platform) | Phased rollout: a subset of cyber defenders first, then trusted partners via the Fairwind Program; developers, businesses and consumers “coming soon” |
| Public model ID | gemini-4-argon (stable) | Not published |
| Context window | 1,048,576 tokens (unified input and output) | Not published |
| Max output | 65,536 default synchronous; up to 1,048,576 with long decode | Not published |
| Inputs / output | Text, image, video, audio, PDF → text | Not published |
| Caching | Implicit caching supported; explicit caching not supported | Not published |
| Official introductory price (per 1M tokens) | $2 input / $10 output; cached input 95% off | Same: $2 input / $10 output; cached input 95% off |
| Headline benchmark claim | Outperforms other frontier models on Google’s cross-domain benchmark set (launch post) | #1 on Artificial Analysis Intelligence Index v5.0 at 59; +89 Elo in web development; 88.2% on SEC-bench Pro |
| On byesu | Yes — gemini-4-argon | Not yet; will be listed after general availability |
What Google Actually Changed
Carbon builds on Argon, released only two weeks earlier. Reading the two launch posts side by side, the deltas Google chose to highlight are:
- Long-horizon engineering. Google says internal teams have had Carbon run for days on multi-step engineering tasks such as root-causing and patching subtle bugs across large codebases, and that it is built for sustained agentic workflows.
- Coding and full-stack web development. Carbon gains +89 Elo on Google’s web-development evaluation, with better quality of output and artifacts.
- Cybersecurity defense. This is the headline of the Carbon post: state-of-the-art results across vulnerability detection, validation and remediation, and more than 2x uplift over Argon in generating secure code on internal benchmarks. It is also the reason for the cautious rollout.
- Alignment and reliability. Google reports fewer unsupported claims (about 2x reduction vs Argon), better behaviour under ambiguity, less overreach and under-service in multi-step plans, and 0.0% attack success on the Gray Swan indirect prompt injection benchmark (Argon was 0.7%). It also reports that successful shortcut-taking on Terminal-Bench 4.0 fell to 0.0%.
What did not change, per Google: the price. Carbon launches at Argon’s introductory price, so when it does reach the API the decision is about capability and behaviour, not cost.
What Developers Should Do Now
- Build on
gemini-4-argon. It is the current flagship and the only Gemini 4 model with a public ID. Nothing in the Carbon announcement deprecates Argon. - Keep the model ID in one place. When Carbon reaches general availability it will arrive as a new model ID. If the ID lives in a single routing config rather than in every call site, switching or A/B-testing is one edit. This is the same lesson as every forced migration — see our GPT-4o shutdown guide for the OpenAI version of it.
- Prepare an evaluation set now. Carbon’s published gains are in long-horizon coding, web development and security review. If those are your workloads, assemble 20–50 real tasks with stored Argon outputs so you can compare the day Carbon is available instead of reading benchmark tables.
- Do not assume caching parity. Argon supports implicit caching only. Google has not published Carbon’s caching behaviour; verify it on launch before porting a cost model.
- Watch the official channels, not announcements of announcements. The Carbon post says access expands through the Fairwind Program and that developer access is coming soon, with no date. Google’s models page is where a public ID will appear first.
Call Gemini 4 Argon With One Key
gemini-4-argon is on byesu with the default auto-group token. The base URL depends on the API format:
- OpenAI-compatible (
/v1/chat/completions,/v1/responses) →https://byesu.com/v1 - Gemini native format →
https://byesu.com
from openai import OpenAI
client = OpenAI(api_key="sk-YOUR_TOKEN", base_url="https://byesu.com/v1")
resp = client.chat.completions.create(
model="gemini-4-argon",
messages=[{"role": "user", "content": "Review this function for concurrency bugs and propose a fix"}],
)
print(resp.usage, resp.choices[0].message.content[:300])The same token also reaches Claude, GPT and Grok models, so a cross-vendor comparison on your own tasks is a loop over model IDs. byesu bills per token at the rate shown in the model list after you sign in, which is byesu’s own rate rather than Google’s list price, so compare the two when you plan a budget.
FAQ
Is Gemini 4 Carbon available in the Gemini API?
Not generally, as of 2026-10-29. Google announced Carbon on 2026-10-28 and said it would first reach a subset of cyber defenders that week, then additional trusted partners in phases through its Fairwind Program, with developer, business and consumer access described as “coming soon”. Gemini 4 Argon has been generally available via the API since 2026-10-13.
Do Gemini 4 Argon and Carbon cost the same?
Yes, at the published introductory price: $2 per million input tokens and $10 per million output tokens, with cached input tokens at 95% off, for both models according to Google. Google labels both as introductory prices, so check the Gemini API pricing page for the current figure.
What is the difference between Gemini 4 Argon and Carbon?
Both are Gemini 4 frontier models. Argon is the generally available flagship reasoning model for complex engineering, math and science, and multi-step agents. Carbon is the newer model Google describes as its strongest, ranked first on the Artificial Analysis Intelligence Index v5.0 at 59, with gains in coding, long-horizon agentic work and cybersecurity defense. Carbon has no public model ID or developer API access yet.
Can I call Gemini 4 Argon through byesu?
Yes. gemini-4-argon is on byesu with the default auto-group token, through the OpenAI-compatible endpoint https://byesu.com/v1 or the Gemini native format at https://byesu.com. Carbon will be listed once Google makes it generally available.
Sources
- Google — Introducing Gemini 4 Argon (2026-10-13): https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-argon/
- Google — Gemini 4 Carbon: our next frontier model (2026-10-28): https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/
- Gemini API — Gemini 4 Argon model page: https://ai.google.dev/gemini-api/docs/models/gemini-4-argon
- Gemini API — Pricing: https://ai.google.dev/gemini-api/docs/pricing