Gemini 4 Argon is now available on the paid Gemini API at an introductory $2 per million input tokens and $10 per million output tokens. The rates rise to $4 and $20 on December 1, 2026. DeepSeek V4.1 Flash and GLM-5.3 list lower token rates, but the right choice depends on tokens consumed, cache hits, tools and the quality your workload needs. The comparison table below uses first-party API prices, not coding subscriptions.
What changed when Google opened Gemini 4 Argon on October 13?
Google announced Gemini 4 Argon on September 30, 2026, with initial access limited to trusted cyber defenders in its Fairwind Program. On October 13, Google announced broader access for developers and enterprises through the Gemini API and AI Studio. This is the new availability and pricing event behind this comparison, rather than the model's first announcement.
Google also brought Argon to eligible subscribers in the Gemini app and other Google products. Those subscriptions have their own access rules. The prices in this article are for the paid Gemini API Standard tier; there is no Argon Free-tier API rate.
What is Gemini 4 Argon, and what work is it meant to handle?
gemini-4-argon is Google's flagship reasoning model for long, multi-step work. Google describes complex software engineering, professional analysis, mathematics and science as target uses. The API documentation lists a shared one-million-token context window, up to one million output tokens and configurable low, medium and high thinking levels.
Google's launch article includes examples from Chrome, business workflows and scientific research. Those are Google-reported uses, not independent SaveMyToken tests. They help explain the intended workload; they do not establish that Argon will finish every task faster or at a lower total cost than a competing model.
Why is the introductory price important?
The $2 input, $10 output and $0.10 cached-input rates run through November 30, 2026. Starting December 1, Google's published Standard rates become $4, $20 and $0.20. Output billing includes thinking tokens. Argon supports implicit prompt caching; its older explicit-cache workflow is not supported. A budget based on the launch rate therefore needs an expiry date and an observed cache-hit rate.
How should you compare Gemini 4 Argon with DeepSeek and GLM?
Hold the workload constant: record uncached input, cached input, output and any paid tool calls for each accepted result. As a simple token-only illustration, one million uncached input tokens plus 200,000 output tokens would cost $4 at Argon's introductory rate, $2.28 on GLM-5.3, or $0.27–$0.54 on DeepSeek V4.1 Flash depending on its off-peak or peak window. This is arithmetic from posted rates, not a measured cost per completed task.
DeepSeek's deepseek-flash endpoint currently serves V4.1 Flash. Its peak windows are 01:00–04:00 and 06:00–10:00 UTC on weekdays, excluding Chinese public holidays; other times use the lower off-peak rates. GLM-5.3 here means Z.ai's direct USD API price. The reference rows below use each provider's direct USD listing, so they do not reuse currency-converted figures from a comparison graphic.
About Claude Code: Claude Code is a coding product, not a single token-priced model. The Claude rows show first-party Claude API rates for Opus 5.5 and Sonnet 5.5. They are not Claude Code subscription prices. The same distinction applies to other coding plans. For a broader shortlist, see the SaveMyToken model comparison.
How do official API prices compare in October 2026?
USD per one million tokens, checked October 14, 2026. Rows compare regular interactive API rates; promotional and regional conditions are noted after the table. A cached-input price applies only when the provider actually records a qualifying cache hit.
| Model / provider | Input | Output | Cached input |
|---|---|---|---|
| Gemini 4 Argon · Google | $2.00 | $10.00 | $0.10 |
| DeepSeek V4.1 Flash | $0.15–$0.30 | $0.60–$1.20 | $0.003–$0.006 |
| GLM-5.3 · Z.ai | $1.40 | $4.40 | $0.26 |
| Muse Wonder 1.0 · Meta | $4.00 | $20.00 | $0.20 |
| Claude Opus 5.5 · Anthropic | $4.00 | $20.00 | $0.20 |
| Claude Sonnet 5.5 · Anthropic | $2.00 | $10.00 | $0.10 |
| GPT-6 Astra · OpenAI | $10.00 | $50.00 | $1.00 |
| GPT-6.1 Sol · OpenAI | $2.00 | $10.00 | $0.10 |
| MiMo V2.6 Pro · Xiaomi | $0.435 | $0.87 | $0.0036 |
| Kimi K3 · Moonshot | $3.00 | $15.00 | $0.30 |
| Mistral Large 4 Preview · Mistral AI (France) | $0.68 | $2.09 | $0.07 |
Want to try a cost-effective Coding Plan featuring DeepSeek V4.1 Flash, GLM-5.3, MiMo V2.6 Pro, and Kimi K3? Visit our coding plan.
- Gemini: introductory Standard rates through November 30; from December 1, input/output/cache become $4/$20/$0.20. Cached input is implicit-cache pricing.
- DeepSeek: ranges show off-peak first, peak second. The API model ID is
deepseek-flash. - Claude, OpenAI and Kimi: cache writes may be charged separately; OpenAI's listed rates use Standard short-context processing. Tools, special processing tiers and subscriptions are excluded.
- MiMo: Xiaomi's overseas, real-time
mimo-v2.6-prorate; domestic CNY, batch and UltraSpeed pricing differ. - Mistral: Large 4 public-preview launch discount is 50% for two weeks from October 6; regular listed input/output/cache prices are $1.36/$4.18/$0.14. Recheck the active price before purchasing.
Sources and price basis
- Google's October 13 Gemini 4 Argon launch announcement ↗ — wider availability and Google-reported uses.
Official API rates were checked October 14, 2026. Rates can change. No cross-model quality or task-cost benchmark was run for this article.