← Back to blog
#IA#Gemini#Google#Programación#APIs

Gemini 4 Carbon Is Now Open to Everyone: The Launch Price Ends November 30

Laptop with a code editor open next to a phone

On November 10, 2026, Google opened Gemini 4 Carbon, its most capable AI model so far, to the general public. It had been announced on October 28 with access restricted to a program for cybersecurity defenders (Fairwind). According to VentureBeat and Diario Qué!, it is now available through the paid API, the Gemini app and Google Antigravity. For developers and teams building with AI, the news is the price: there is a launch rate with an expiry date.

What we know about the launch

  • Availability: through the paid Gemini API and Gemini for business. For consumers, for Google AI Ultra subscribers in the Gemini app, AI Mode in Search and Antigravity (in the US the plan costs USD 99.99 or USD 199.99 a month, per VentureBeat). Google hasn't published a country-by-country schedule, so check your own account to see whether your region has it.
  • It's a hosted model: there are no downloadable weights or license to run it on your own servers.
  • Positioning: according to Diario Qué!, Google recommends continuing to use the previous model (Argon) for most tasks and reserving Carbon for complex coding, debugging and harder problems.

Pricing: a promotion that ends November 30

Per million tokens, Carbon costs USD 2 input and USD 10 output through November 30, 2026; afterwards it rises to USD 4 and USD 20. According to VentureBeat, Claude Opus 5.5 costs USD 4 and USD 20 at standard rates: during the promotion, through November 30, Carbon costs half as much; from December 1, the same. The official Gemini API changelog confirms general availability on November 10 and the USD 2 and USD 10 rate through November 30; the later USD 4 and USD 20 rate is reported by VentureBeat (the changelog points to the pricing page). The comparison with other Anthropic models comes from a single source: check each provider's current rates before budgeting.

The practical point: if you estimate costs with the launch rate, your December budget will double. Do the math with USD 4 and USD 20.

Benchmarks, with caution

Google claims Carbon leads several coding and automation tests (CursorBench, WANDR, Zapier's AutomationBench) and security tests (SEC-bench Pro, ExploitGym, CWE-bench). VentureBeat warns that part of those claims can't be verified with public data and cites an independent Artificial Analysis index in which Carbon (60) ranks behind Claude Fable 5.2 (65) and ahead of Claude Opus 5.5 (59). On other tests the order changes depending on who publishes them. The sensible conclusion: vendor benchmarks guide you, but they don't replace your own test.

What this means for your SME and your technical team

  • For owners and executives: competition among Google, Anthropic and OpenAI keeps lowering the cost per task. It's a good time to automate customer service, quotes or reports; we covered the trend in the AI model price war.
  • For developers: test Carbon with your own code and cases before migrating. Measure three things: quality of the output, cost per completed task and response time. A model that is cheaper per token can end up costing more if it needs more attempts.
  • For students: the most powerful access is paid; to learn, free plans and earlier models remain enough. See these tools for developers.

Four precautions before integrating it

  1. Don't marry a single provider. Wrap model calls in a layer of your own so you can switch models if price or quality changes. It's the risk we describe in AI wrappers and real value.
  2. Personal data. If the model will process customer or employee data, review what is sent, where it is processed and what the provider's contract says. Ecuador's personal data protection rules apply.
  3. Set spending caps and alerts on the API account from day one.
  4. Version your tests. Keep a set of cases to rerun whenever the model or its price changes.

How we approach it at SimCodec

At SimCodec we design AI solutions (voice and WhatsApp agents, process automation and custom software) without tying the client to a single model: we measure with real cases from your operation and choose the model with the best quality-to-cost ratio that month. If you want to assess which model fits your case, contact us.

Sources

← Back to blog Get a quote →