Gemma 4 12B IT vs Gemma 3 12B API pricing
Updated
Gemma 4 12B IT: USD 0.10 input, USD 0.30 output per 1M tokens (SiliconFlow). Gemma 3 12B: USD 0.05 input, USD 0.15 output per 1M tokens (DeepInfra and 1 other). At 3 input tokens per output token, Gemma 4 12B IT costs 2× as much as Gemma 3 12B (USD 0.15 vs USD 0.075 per 1M).
How do Gemma 4 12B IT and Gemma 3 12B differ on price?
- Cheapest input price: Gemma 3 12B USD 0.05 against USD 0.10 per 1M tokens, so Gemma 4 12B IT costs 2× as much.
- Cheapest output price: Gemma 3 12B USD 0.15 against USD 0.30 per 1M tokens, so Gemma 4 12B IT costs 2× as much.
- Among the 201 indexed models, Gemma 4 12B IT ranks 40th cheapest and Gemma 3 12B 11th on the 3:1 blend.
- No input:output mix changes the answer: Gemma 3 12B is cheaper at every mix before cache discounts.
- Only Gemma 4 12B IT lists a cached-input discount: Gemma 4 12B IT at 1 standard price of 2, from USD 0.15 (Sail Research, 50% below that provider's input price). Gemma 3 12B's cached tokens bill at the input price.
- Gemma 4 12B IT states up to 262,144 tokens of context, 2× the 131,072 of Gemma 3 12B.
- Gemma 3 12B has 4 providers with a standard price, 2 more than Gemma 4 12B IT; no provider sells both, so using both means two accounts.
- Shopping around matters more for Gemma 3 12B: its dearest standard price is 5.3× its cheapest, against 4.8× for Gemma 4 12B IT.
- For the one-developer coding agent preset, the gap is USD 1,242 a year in Gemma 3 12B's favour (Gemma 4 12B IT USD 207.00 a month, Gemma 3 12B USD 103.50).
Gemma 4 12B IT and Gemma 3 12B prices side by side
USD per 1M tokens, standard tier. "Cheapest" is the lowest price at 3 input tokens to 1 output token; each line can come from a different provider.
| Price | Gemma 4 12B IT | Gemma 3 12B |
|---|---|---|
| Model page | Gemma 4 12B IT Google · 2 providers | Gemma 3 12B Google · 4 providers |
| Cheapest standard price, blended 3:1 | USD 0.15 USD 0.10 in / USD 0.30 out at SiliconFlow SiliconFlow page checked 2026-11-14 | USD 0.075 lower USD 0.05 in / USD 0.15 out at DeepInfra and 1 other DeepInfra page checked 2026-11-14 |
| Cheapest input | USD 0.10 SiliconFlow source | USD 0.05 lower DeepInfra source |
| Cheapest cached input | USD 0.15 Sail Research source | No standard price lists a cache price; cached tokens bill at the input price. |
| Cheapest output | USD 0.30 SiliconFlow source | USD 0.15 lower DeepInfra source |
| Maker's own API | Google publishes no paid standard price on its own API that we found. | Google publishes no paid standard price on its own API that we found. |
| Median standard price | USD 0.20 / USD 1.15 input / output, 2 prices | USD 0.07 / USD 0.22 input / output, 4 prices |
| Context window | 262,144 tokens, largest stated | 131,072 tokens, largest stated; from 32,768 at some providers |
| Providers with a standard price | 2 | 4 |
- Lower marks the lower price on a line; tie, equal prices. Other tiers and free rows are on the model pages.
Report a wrong price or number (email to info@broadnet.ai, with this page's link filled in).
What do Gemma 4 12B IT and Gemma 3 12B cost for a real workload?
30 days at each model's cheapest standard price for the workload (presets of the cost calculator; cached input at the cached price where listed).
| Workload | Gemma 4 12B IT | Gemma 3 12B | Difference |
|---|---|---|---|
| Support chatbot 2,000 requests a day, 1,500 input tokens (60% cached), 300 output | USD 14.40 SiliconFlow | USD 7.20 DeepInfra | USD 7.20 Gemma 3 12B lower |
| RAG over documents 1,000 requests a day, 8,000 input tokens (30% cached), 500 output | USD 28.50 SiliconFlow | USD 14.25 DeepInfra | USD 14.25 Gemma 3 12B lower |
| Coding agent (one developer) 1,500 requests a day, 40,000 input tokens (85% cached), 2,000 output incl. reasoning | USD 207.00 SiliconFlow | USD 103.50 DeepInfra | USD 103.50 Gemma 3 12B lower |
Gemma 3 12B is cheaper on all three workloads. The widest gap is on Support chatbot: the dearer model costs 2× as much as the cheaper one.
Does the cheaper model change with your input:output ratio?
Gemma 3 12B is cheaper at every mix of input and output tokens, before cache discounts: its cheapest input price and its cheapest output price are both at or below Gemma 4 12B IT's, so the input:output ratio never changes the answer.
| Input : output tokens | Gemma 4 12B IT | Gemma 3 12B |
|---|---|---|
| 1 : 3 (long answers) | USD 0.25 SiliconFlow | USD 0.125 lower DeepInfra |
| 1 : 1 | USD 0.20 SiliconFlow | USD 0.10 lower DeepInfra |
| 3 : 1 (the site’s blend) | USD 0.15 SiliconFlow | USD 0.075 lower DeepInfra |
| 10 : 1 (retrieval, long prompts) | USD 0.1182 SiliconFlow | USD 0.05909 lower DeepInfra |
| 30 : 1 (agents, big contexts) | USD 0.1065 SiliconFlow | USD 0.05323 lower DeepInfra |
USD per 1M tokens at the cheapest standard provider for each mix, before cache discounts. Thinking tokens bill as output.
What do Gemma 4 12B IT and Gemma 3 12B cost per character in other languages?
Not measured: the real cost by language dataset covers only tokenizers we could measure from published files, and Gemma 4 12B IT is not among them. Count tokens on your own text with the counters below.
Check it on your own traffic
- Gemma 4 12B IT API pricing: all 2 providers and other tiers.
- Gemma 3 12B API pricing: all 4 providers and other tiers.
- Cost calculator with Gemma 4 12B IT and Gemma 3 12B preloaded
- Gemma token counter
- Gemma 3 token counter
- Prompt caching calculator
More comparisons: Gemma 4 26B A4B vs Gemma 4 12B IT · Gemma 3 27B vs Gemma 3 12B · Gemma 3 12B vs Gemma 3 4B · every Gemma 4 12B IT comparison · every Gemma 3 12B comparison · all 662 price comparisons
Questions
Is Gemma 4 12B IT cheaper than Gemma 3 12B?
No. On a blend of 3 input tokens to 1 output token, Gemma 3 12B costs USD 0.075 per 1M tokens and Gemma 4 12B IT USD 0.15, at each model's cheapest standard price. For input alone, Gemma 3 12B is cheaper (USD 0.05 against USD 0.10 per 1M); for output alone, Gemma 3 12B is (USD 0.15 against USD 0.30). This compares price only, not quality.
What do Gemma 4 12B IT and Gemma 3 12B cost per 1M tokens?
Gemma 4 12B IT: from USD 0.10 input and USD 0.30 output at SiliconFlow; median of 2 standard prices USD 0.20 / USD 1.15. Gemma 3 12B: from USD 0.05 input and USD 0.15 output at DeepInfra and 1 other; median of 4 standard prices USD 0.07 / USD 0.22. Each price is read from the provider's own pricing page; the source is linked in the table.
Which providers sell both Gemma 4 12B IT and Gemma 3 12B?
None of the providers we track sells both at a standard price. Gemma 4 12B IT is cheapest at SiliconFlow and Gemma 3 12B at DeepInfra and 1 other, so running both means two accounts.
Which is cheaper for a coding agent or a chatbot, Gemma 4 12B IT or Gemma 3 12B?
For a coding agent (1,500 requests a day, 40,000 input tokens (85% cached), 2,000 output incl. reasoning): USD 207.00 a month on Gemma 4 12B IT (SiliconFlow) against USD 103.50 on Gemma 3 12B (DeepInfra), a difference of USD 103.50. For a support chatbot (2,000 requests a day, 1,500 input tokens (60% cached), 300 output): USD 14.40 a month on Gemma 4 12B IT (SiliconFlow) against USD 7.20 on Gemma 3 12B (DeepInfra), a difference of USD 7.20. Each uses the cheapest standard price for that workload, cached input included. Try your own numbers.
At what input:output ratio does the cheaper of Gemma 4 12B IT and Gemma 3 12B change?
Gemma 3 12B is cheaper at every mix of input and output tokens, before cache discounts: its cheapest input price and its cheapest output price are both at or below Gemma 4 12B IT's, so the input:output ratio never changes the answer.
Do Gemma 4 12B IT and Gemma 3 12B have cached input pricing?
Gemma 4 12B IT from USD 0.15 per 1M cached input tokens (Sail Research; 1 of 2 standard prices list a discount). Gemma 3 12B: no standard price lists a cached-input discount. Cached tokens are billed at the input price where no cache price is published.
Which has the larger context window, Gemma 4 12B IT or Gemma 3 12B?
Gemma 4 12B IT: up to 262,144 tokens, against 131,072 for Gemma 3 12B, as stated on providers' pricing pages. Some providers offer less than the maximum; check the model pages before you pick the cheapest row for long prompts.