Gemma 3 12B
GoogleReleased Mar 13, 2025Open weights · Gemma Terms of Usegemma-3-12b
- Intelligence (Epoch)
- 124 · 101st of 109
- Reasoning
- no
- Knowledge cutoff
- January 2025
- Context
- 128k tokens
- Max output
- 16k tokens
- Input
- text, image
- API features
- tools, structured outputs, JSON mode, seed
Read Nov 3, 14:34 UTC from the hosts' published offers and OpenRouter, spec sheet from Google · cutoff: ai.google.dev
The cheapest host, Novita AI, does not state its weights precision. Cheapest at the maker's precision and context: DeepInfra, $0.20 blended. Weights and context per host
| Provider | Input | Output | Cache ↑ | tok/s | First token | Uptime 24h | Checked |
|---|---|---|---|---|---|---|---|
| NextBitint4 · 128K ctx | $0.075 | $0.15 | $0.05 | 55 | 0.99 s | 100% | Nov 3 |
| Novita AIprecision not stated · 128K ctx · 8K out | $0.05 | $0.10 | — | — | — | — | Nov 3 |
| DeepInfrabf16 · 128K ctx · 16K out | $0.05 | $0.15 | — | 22 | 0.87 s | 93.0% | Nov 3 |
3 of 3 providers · prices per million tokens, read Nov 3, 14:34 UTC · speed: OpenRouter, median of the last 30 min, read Nov 3, 14:34 UTC · uptime: OpenRouter, last 24 h · under each host: weights precision, context and max output as served, the maker's value in brackets when the host's differs
| Provider | Weights | Context | Max output | Tools | Structured outputs | Reasoning |
|---|---|---|---|---|---|---|
| Novita AI | not stated by the host | 128K | 8K | — | — | — |
| DeepInfra | bf16 | 128K | 16K | yes | yes | no |
| NextBit | int4 | 128K | not read: host terms | no | yes | no |
Weights, context, max output: from the host where its terms allow, else as declared to OpenRouter (hover for source and date) · features as declared to OpenRouter, not verified by us
Gemma 3 12B spec sheet, field by field, with sources
Gemma 3 12B spec sheetmaker-declared
What Google declares, from its own weights and docs. What each host serves is in the offers: both can be true.
- Context window
- weights repository under access conditions
- Parameters
- weights repository under access conditions
- Active experts
- weights repository under access conditions
- Weight precision
- weights repository under access conditions
- License
- Gemma Terms of Use — open weights, conditionalcuratedScope: the license of the published weights only — not the maker's acceptable-use policy, nor the terms of the host you call.
- Knowledge cutoff
- January 2025maker's docs
- Modalities
- weights repository under access conditions
- Max output (official)
- not published by the makermaker's docs · read 2026-10-23
- Release date
- to verify≈ March 13, 2025 third-party catalogue — not the maker's figure
- Maker lifecycle status
- to verify
google/gemma-3-12b-it · weights repository @ 96b6f1e · file not readable · read 2026-10-08 · 2 of 10 fields published by the maker
| Date | Provider | Input | Output | Change |
|---|---|---|---|---|
| Oct 23 | NextBit | $0.10 → $0.075 | $0.17 → $0.15 | −18% |
| Oct 21 | NextBit | $0.05 → $0.10 | $0.15 → $0.17 | +38% |
Every published change of input or output price, newest first · change on input + output · full series: pricing-history.csv.
| Provider | Chatbot 1,000 in / 500 out | RAG / long context 8,000 in / 500 out | Batch summarize 4,000 in / 1,000 out |
|---|---|---|---|
| Novita AI | $0.10 | $0.45 | $0.30 |
| DeepInfra | $0.12 | $0.47 | $0.35 |
| NextBit | $0.15 | $0.68 | $0.45 |
What each host bills for 1,000 requests of a representative job
Your own workload
Intelligence for the price
1 model scores higher for the same price or less; it ranks 2 of the 5 models at its price or below.
Epoch Capabilities Index against the lowest input + output price per M, log scale · prices Nov 3. Score: Epoch AI, CC BY 4.0, 124 with an interval of 117 to 128.
Against its rivals
The best of its price range and the top score
- Gemma 3 12Bthis model$0.15124
- GPT-6 Astratop score$60.00169
- DeepSeek V4.1 Flashsame price range$0.17156
- DeepSeek V4 Flash 0731same price range$0.21155
Same level for 3.9× less: Mistral Nemo at Io Net, $0.038 per M input + output · their Epoch intervals overlap (119 against 124).
Bars: Epoch Capabilities Index, drawn from 100 · price: lowest input + output per M.
Run it yourself
open weights, Gemma Terms of Use · fits 14 of the 14 workstations in our guideSelf-host guide ›- Parameters
- 12B
- dense
- Memory needed
- 7 GB
- in INT4 · 14 GB in INT8
- Smallest cloud setup
- RTX 3090
- 24 GB · $0.18/h, about $131 a month
- Worth it above
- 876M tokens
- a month, against the cheapest API at $0.15 per M
| Quantization | Memory | Smallest cloud GPU | Quality |
|---|---|---|---|
| INT4 (Q4) | 7 GB | RTX 3090 · $0.18/h | smallest footprint, minor quality loss |
| INT8 (Q8) | 14 GB | RTX 3090 · $0.18/h | near-lossless |
| FP16 (full) | 29 GB | A100 · $1.19/h | reference quality |
At home: in INT4 it fits the Intel Arc Pro B60 (24 GB) + host PC — about $1,666 once (host PC included), or $0.0634/h over 3 years, electricity aside, at 456 GB/s of memory bandwidth — that, not FLOPS, sets tokens per second; it pays for itself after about 11B tokens against the cheapest API.
Memory ≈ parameters × bytes × 1.2 · cloud: on-demand, 24/7 · builds: GGUF, AWQ, MLX, GPTQ, Ollama · break-even calculator
Other Google models · all its models, hosts and history ›
- Gemma 4 31B14 providers · from $0.44
- Gemma 4 26B A4B13 providers · from $0.31
- Gemma 3 27B5 providers · from $0.24
- Gemma 3 4B2 providers · from $0.09
- Gemini 2.5 Flash1 provider · from $2.80
- Gemini 2.5 Pro1 provider · from $11.25