comparisons

Gemini 4 Carbon API Pricing: PoYo vs Google, Cost per Request

Poyo.ai Team
9 min read
Share:

Gemini 4 Carbon API pricing on PoYo compared with Google

Gemini 4 Carbon API pricing on PoYo is $1.60 per 1M input tokens, $8.00 per 1M output tokens, and $0.08 per 1M cache-read tokens for the model ID gemini-4-carbon, which is 20% below Google's introductory input and output rates. Google's own price doubles on December 1, 2026. This post puts both price lists side by side and works out the cost of three realistic request shapes. Current PoYo rates are always on the Gemini 4 Carbon model page.

PoYo prices below are as listed on November 16, 2026. Google prices come from its Gemini API pricing page, read the same day.

Gemini 4 Carbon API price per 1M tokens

Meter (per 1M tokens)PoYoPoYo creditsGoogle standard, through Nov 30, 2026Google standard, from Dec 1, 2026
Input$1.60320$2.00$4.00
Output, incl. thinking tokens$8.001,600$10.00$20.00
Cache read$0.0816$0.10$0.20

One PoYo credit is $0.005, so 320 credits = $1.60. Google introduced Carbon on October 28, 2026 at the same introductory rate as Gemini 4 Argon, with cached input at 95% off, and moved it to the Gemini API paid tier on November 10, 2026. Google says the $4 / $20 price applies once "the introductory period expires on November 30, 2026."

Google also publishes batch prices for Carbon ($1.00 input / $5.00 output through November 30, then $2.00 / $10.00). Those apply to Google's Batch API. The PoYo prices here are for the synchronous endpoints: POST /v1/chat/completions and POST /v1beta/models/gemini-4-carbon:generateContent / :streamGenerateContent.

What is and is not billed

ItemBilled?Source
Prompt tokens (text and media)Yes, input ratePoYo price list
Visible answer tokensYes, output ratePoYo price list
Thinking tokensYes, at the output rateGoogle pricing page: thinking tokens are included in the output price
Prompt tokens served from cacheYes, cache-read rate ($0.08 per 1M on PoYo)PoYo price list
Explicit cache creation / storageNot applicable: Google lists implicit caching only, "Explicit caching is not supported"Google pricing and model pages
Free tierNone at Google: "not available on the Free tier"Google pricing page
Google Search / Maps groundingGoogle charges $14 per 1,000 requests after 5,000 free per month; PoYo's documented request schema has no grounding fieldGoogle pricing page, PoYo docs

Because caching is implicit, you cannot force a cache hit; you can only make one likely by keeping a long, identical prefix at the start of each request. Read usage from usage (Chat Completions) or usageMetadata (Gemini Native). The field-by-field mapping is in the Gemini 4 Carbon API guide.

Worked Gemini 4 Carbon API cost examples

All three examples use token counts you can plug your own numbers into. Output counts include thinking tokens.

1. Repository review: 200K input, 8K output

One request with a 200,000-token slice of a codebase and an 8,000-token review.

Price listArithmeticCost
PoYo0.2 × $1.60 + 0.008 × $8.00 = $0.32 + $0.064$0.384 (76.8 credits)
Google, through Nov 300.2 × $2.00 + 0.008 × $10.00 = $0.40 + $0.08$0.48
Google, from Dec 10.2 × $4.00 + 0.008 × $20.00 = $0.80 + $0.16$0.96

At 1,000 reviews a month that is $384 on PoYo versus $480 (intro) or $960 (from December 1) at Google list price.

2. Agent loop: 20 turns with a growing context

An agent starts with a 30,000-token prompt (system prompt, tool definitions, task). Each turn adds 5,000 tokens of tool results and prior output, and the model writes 2,000 output tokens per turn.

input per turn t  = 30,000 + 5,000 × (t − 1)
total input       = 20 × 30,000 + 5,000 × (0 + 1 + … + 19)
                  = 600,000 + 950,000 = 1,550,000 tokens
total output      = 20 × 2,000 = 40,000 tokens
CaseArithmeticPoYoGoogle, through Nov 30
No cache hits1.55 × $1.60 + 0.04 × $8.00$2.80 (560 credits)$3.50
Best case: every earlier turn read from cache0.125 × $1.60 + 1.425 × $0.08 + 0.04 × $8.00$0.634 (126.8 credits)$0.7925

In the best case only the first 30K tokens plus 19 × 5K new tokens (125,000) are billed at the input rate; the other 1,425,000 are cache reads. Real runs land between the two rows, because implicit caching is not guaranteed. Either way, the input side dominates long loops, so stable prefixes save more than trimming output. Agent-specific patterns are in Gemini 4 Carbon API for coding agents.

3. Cached long-document Q&A: 400K document, 50 questions

A 400,000-token contract bundle is the fixed prefix; each of 50 questions adds 500 tokens and gets a 1,500-token answer. Assume the first call is uncached and the next 49 hit the cache.

input rate:  400,500 (first call) + 49 × 500 = 425,000 tokens
cache read:  49 × 400,000                   = 19,600,000 tokens
output:      50 × 1,500                     = 75,000 tokens
Price listArithmeticCost
PoYo, with cache hits0.425 × $1.60 + 19.6 × $0.08 + 0.075 × $8.00 = $0.68 + $1.568 + $0.60$2.848 (569.6 credits)
PoYo, no cache hits20.025 × $1.60 + 0.075 × $8.00$32.64
Google intro, with cache hits0.425 × $2.00 + 19.6 × $0.10 + 0.075 × $10.00$3.56

Cache hits cut this workload by more than 90%. Keep the document first and the question last so the prefix stays identical.

Gemini 4 Carbon vs Gemini 4 Argon API price

Meter (per 1M, PoYo)gemini-4-carbongemini-4-argon
Input$1.60$1.60
Output$8.00$8.00
Cache read$0.08$0.16

Per token, Carbon and Argon cost the same for input and output on PoYo and at Google. On PoYo, Carbon's cache read is half of Argon's, so example 3 on Argon would cost 0.425 × $1.60 + 19.6 × $0.16 + 0.075 × $8.00 = $4.416 with the same token counts.

The catch is token counts per task. Google says Carbon works by "using more tokens than Argon to explore thoroughly, reason deeply, and run extended verification loops," and that Argon "remains the ideal choice for the majority of software engineering and knowledge work." Google does not publish a multiplier, so measure output tokens on your own prompts with both IDs before you move traffic. See Gemini 4 Carbon API vs Gemini 4 Argon and the Gemini 4 Argon API guide.

Gemini 4 Carbon API vs Claude Opus 5.5 and GPT-6 Astra on PoYo

All prices below are PoYo list prices per 1M tokens:

Model IDInputOutputCache readExample 1 (200K in / 8K out)
gemini-4-carbon$1.60$8.00$0.08$0.384
claude-opus-5-5$3.20$16.00$0.16$0.768
gpt-6-astra$8.00$40.00$0.80$1.92

Claude Opus 5.5 also bills cache writes on PoYo ($4.00 per 1M for 5-minute and $6.40 per 1M for 1-hour), and GPT-6 Astra bills cache writes at $10.00 per 1M. Google lists no explicit cache write for Carbon. The example column assumes the same token counts on every model, which is a price comparison, not a cost-per-task claim: each model spends a different number of tokens on the same job. For benchmark-based routing between the three, read Gemini 4 Carbon API vs Claude Opus 5.5 vs GPT-6 Astra. The model overview is in Gemini 4 Carbon API.

Frequently asked questions

How much does the Gemini 4 Carbon API cost per 1M tokens?

On PoYo: $1.60 input, $8.00 output, $0.08 cache read (320 / 1,600 / 16 credits). Google: $2.00 / $10.00 / $0.10 through November 30, 2026, then $4.00 / $20.00 / $0.20.

Are Gemini 4 Carbon API thinking tokens billed?

Yes. Google's pricing page says thinking tokens are included in the output price, so they are billed at the output rate ($8.00 per 1M on PoYo).

Is the Gemini 4 Carbon API cheaper than Gemini 4 Argon?

Per token, input and output prices are the same; Carbon's PoYo cache-read price is lower ($0.08 vs $0.16). Per task, Carbon can cost more because Google says it uses more tokens than Argon.

Does the Gemini 4 Carbon API have a free tier?

No. Google's pricing page says Carbon is not available on the free tier. PoYo is pay-as-you-go with credits.

How does Gemini 4 Carbon API caching work?

Google supports implicit caching only, with no explicit cache objects. Repeated prompt prefixes that hit the cache are billed at $0.08 per 1M on PoYo instead of $1.60.

Will the Gemini 4 Carbon API price change after November 30, 2026?

Google's price rises to $4.00 input / $20.00 output on December 1, 2026. Check the model page for PoYo's current rate.

Share: