Gemini 4 Argon vs Carbon
Same price, different thinking budget. Both list at $2 / $10 per 1M input/output tokens through 30 November 2026 and $4 / $20 from 1 December 2026, with the same 1M unified context window and tools. Google says the main difference is that Carbon spends more thinking tokens per turn. In Google’s Carbon model card, Carbon scores higher than Argon on all 17 of 17 benchmarks listed. In Artificial Analysis’s runs at high, Carbon used about 1.8× Argon’s output tokens and cost $5.48 vs $4.97 per Index task.
Source: Google DeepMind, results as of October 2026, retrieved 14 November 2026 · Source: Google AI Studio / Gemini API docs, Retrieved 13 November 2026 · Source: Google AI Studio / Gemini API docs, Retrieved 13 November 2026 · Source: Google AI Studio / Gemini API docs, Retrieved 13 November 2026
Official specs side by side
AI Studio “what’s new” docs for each model and the Gemini API pricing page, retrieved 14 November 2026.
| Gemini 4 Argon | Gemini 4 Carbon |
|---|
| Model ID | gemini-4-argon | gemini-4-carbon |
|---|
| Google’s description | “Efficient Gemini 4 frontier reasoning model”, faster responses and lower token spend per turn | “Strongest reasoning model”, spends more thinking tokens to explore and verify |
|---|
| Price / 1M (in / out) | $2 / $10 through 30 Nov 2026; $4 / $20 after | $2 / $10 through 30 Nov 2026; $4 / $20 after |
|---|
| Thinking levels | low, medium (default), high | low, medium (default), high |
|---|
| Context | 1,048,576-token unified window; long decode up to 1M | 1,048,576-token unified window; long decode up to 1M |
|---|
| Free API tier | No (paid tier) | No (paid tier) |
|---|
| Google says use it for | Interactive coding, day-to-day feature work, fast production loops | Root-cause debugging, architecture, dense multimodal synthesis, proofs, security audits |
|---|
Google’s benchmark table: Carbon vs Argon
The Carbon card also lists GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.1; see Carbon vs Opus 5.5. Note the card’s “AA Index v5.0” (Carbon 59, Argon 50) is Google’s October snapshot; Artificial Analysis’s live page now shows Carbon 60 at high.
Gemini 4 Carbon model card, Google DeepMind, results as of October 2026 (Google-run; cybersecurity via the Fairwind Program). Not reproduced by this site. Bold = higher of the columns shown.| Category | Benchmark | Notes | Gemini 4 Argon | Gemini 4 Carbon |
|---|
| General | AA Index v5.0 | | 50 | 59 |
|---|
| General | Vals Index v2.2 | | 69.2% | 70.4% |
|---|
| Agentic coding | SWE-sweep | | 4.3% | 9.6% |
|---|
| Agentic coding | FrontierSWE 2 | | 55.0% | 61.2% |
|---|
| Agentic coding | Terminal-Bench 4.0 | AA | 57.1% | 65.2% |
|---|
| Computer use | Agents' Last Exam v1.1 | Score | 69.0% | 77.4% |
|---|
| Computer use | OSWorld-2.1 | Offline subset | 74.7% | 81.5% |
|---|
| Knowledge work | AutomationBench | Zapier | 51.3% | 53.9% |
|---|
| Knowledge work | Vending-Bench 2 | Bank balance | $13,718 | $16,484 |
|---|
| Multimodal | Chartography | | 71.6% | 72.5% |
|---|
| Multimodal | Blueprint-Bench 2 | Similarity score | 54.4% | 58.6% |
|---|
| Long context | GraphWalks 256k–1M | BFS (F1) | 84.2% | 96.2% |
|---|
| Science and math | RiemannBench | | 76.0% | 80.0% |
|---|
| Science and math | Terminal-Bench-Science 0.1 | AA | 52.9% | 64.8% |
|---|
| Cybersecurity | CWE-bench | | 68.0% | 71.0% |
|---|
| Cybersecurity | ExploitGym | | 42.0% | 54.8% |
|---|
| Cybersecurity | SEC-bench Pro | | 83.4% | 88.2% |
|---|
Speed and cost per task (Artificial Analysis)
Measured independently by Artificial Analysis on each vendor's own API (retrieved 14 November 2026); speed is averaged over the past 5 days and changes. A dash means AA had not published that figure. This site did not run these tests.| Model · thinking/effort level | AA Intelligence Index | Output speed (tok/s) | Minutes per AA Index task | Time to first answer token (s) |
|---|
| Gemini 4 Argon · high | 50 | 66 | 29.2 | 44.99 |
|---|
| Gemini 4 Argon · medium | 48 | 56 | 29.9 | 6.34 |
|---|
| Gemini 4 Carbon · high | 60 | 66 | 30.9 | 70.53 |
|---|
| Gemini 4 Carbon · medium | 58 | 61 | 27.9 | 22.18 |
|---|
| Gemini 4 Carbon · low | 54 | 51 | 28.0 | 13.89 |
|---|
The task-cost columns are our arithmetic on official list prices, assuming every model uses the same token counts — they do not (see output volume below), and Anthropic says its newer tokenizer produces about 30% more tokens for the same text. No cache, batch, or grounding fees. The last column is Artificial Analysis’s own measured average, at the prices AA used, for the variant named in brackets.| Model | Official list price (input / output per 1M) | Our arithmetic per task (50K in + 8K out) | ×1,000 tasks | AA measured cost per Index task |
|---|
| Gemini 4 Argon | $2 / $10 through 30 Nov 2026; $4 / $20 from 1 Dec 2026 | $0.180 (to 30 Nov 2026)
$0.360 (from 1 Dec 2026) | $180
$360 | $4.97 (high) |
|---|
| Gemini 4 Carbon | $2 / $10 through 30 Nov 2026; $4 / $20 from 1 Dec 2026 | $0.180 (to 30 Nov 2026)
$0.360 (from 1 Dec 2026) | $180
$360 | $5.48 (high) |
|---|
Output tokens Artificial Analysis recorded while running its whole Intelligence Index (more tokens = more cost and time for the same work): Gemini 4 Argon: high 113M, medium 81.7M; Gemini 4 Carbon: high 205M, medium 124M, low 83.9M.
Source: Artificial Analysis: Gemini 4 Argon · Artificial Analysis: Gemini 4 Carbon (retrieved 14 November 2026)
Which to pick (only where official numbers decide it)
- Same bill per token, different bill per task. Per-token prices are identical, so cost differences come from token use. Artificial Analysis measured Carbon at 205M output tokens vs Argon 113M across its Index at high.
- Latency: at high, AA measured time to first answer token of about 70.5 s for Carbon vs 45.0 s for Argon; output speed was similar (66 tok/s each).
- Hardest tasks: the biggest gaps in Google’s table are long context (GraphWalks 96.2% vs 84.2%), ExploitGym (54.8% vs 42.0%) and Terminal-Bench-Science (64.8% vs 52.9%).
- Cheaper than both: Gemini 4 Hydrogen lists at $1.50 / $7.50 and has a free API tier; see Hydrogen vs Argon / Carbon.
- Price change: the introductory $2 / $10 ends on 30 November 2026 for both. Budget at $4 / $20 for December onward.
More comparisons Gemini 4 Carbon vs Opus 5.5 Hydrogen vs Argon / Carbon Hydrogen vs Opus 5.5 All comparisons Cost calculator Pricing
