Up to date
google Gemma Terms of Use out 20 months ago
4.30B parameters, images and text in, text out, 128K tokens of context. 3 providers serve it from $0.040 per million input tokens.
Verified 2026-10-25 Rank #29 in LLMs · score 113.7 method v1 Sources 5 Downloads 30d 988.1k · #111 of 619 tracked
Changed this week: Cheapest input price fell 20%: $0.050 → $0.040 per 1M (OpenRouter).2026-10-24 · 1 new quantized build on the Hub (492 in all).2026-10-24
Picked by rule from 35 tracked pieces: the lab's own post, the most shared, a field report, the latest.
An open-source implementation of Jev option scoring runs on local Gemma 3 4B models using MLX on Apple silicon or PyTorch on other platforms.
The 2 picks above, plus these 23. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.
Live hosted endpoints, cheapest first, USD per million tokens. Ties break on output price, then provider name.
2 providers with a live price, from $0.040 to $0.050 per million input tokens; the median is $0.045. One dot per provider, cheapest in teal.
Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.
Served by 3 API providers; heavily run locally (492 quantized builds); a big base for builders (1,354 fine-tunes, adapters and merges).
Top Spaces running it: ginigen/VEO3-Free ♥ 461 · ginigen/VEO3-Directors ♥ 92 · hadadrjt/ai ♥ 86 · TIGER-Lab/MEGA-Bench ♥ 51
Weights only, an estimate: leave room for the KV cache. Method
About 2.6 GB at INT4, 5.2 GB at FP8, 10 GB at BF16: weights only, so leave room for the KV cache.
Estimate params × bytes per param × 1.2 · how we estimate
Third-party measurements only; author-reported numbers stay in Reference, marked as such.
19.62. Each tick is a model on this site with a measured BFCL row; author-reported numbers stay in Reference. Percent (overall accuracy). Prompt
Source · leaderboard · 2026-10-25
Claim-matched tests, graded by code. Receipts, not a score.
{
"providers": [
{
"provider": "parasail",
"endpoint_model": "parasail-googlegemma-3-4b-it",
"source_url": "https://api.parasail.io/v1/chat/completions",
"ttft_ms_median": 460,
"ttft_any_ms_median": 460,
"tokens_per_s_median": 108.5,
"runs": [
{
"ttft_s": 0.516,
"ttft_any_s": 0.516,
"tokens_per_s": 106.5,
"latency_s": 1.65,
"tokens_out": 176,
"reasoning_chars": 0,
"finish": "stop"
},
{
"ttft_s": 0.454,
"ttft_any_s": 0.454,
"tokens_per_s": 116.6,
"latency_s": 1.51,
"tokens_out": 176,
"reasoning_chars": 0,
"finish": "stop"
},
{
"ttft_s": 0.46,
"ttft_any_s": 0.46,
"tokens_per_s": 108.5,
"latency_s": 1.62,
"tokens_out": 176,
"reasoning_chars": 0,
"finish": "stop"
}
]
}
],
"note": "Single client, wall clock; tok/s counts every output token including hidden reasoning."
} {
"fields_matched": 59,
"fields_total": 80,
"cost_per_1000_usd": 0.443,
"parsed": true,
"per_item": [
{
"street": true,
"city": true,
"postcode": true,
"country": false
},
{
"street": true,
"city": true,
"postcode": true,
"country": false
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": false
},
{
"street": true,
"city": true,
"postcode": true,
"country": false
},
{
"street": true,
"city": true,
"postcode": true,
"country": false
},
{
"street": false,
"city": true,
"postcode": true,
"country": false
},
{
"street": true,
"city": true,
"postcode": true,
"country": false
},
{
"street": true,
"city": true,
"postcode": true,
"country": false
},
{
"street": true,
"city": true,
"postcode": true,
"country": false
},
{
"street": true,
"city": true,
"postcode": true,
"country": false
},
{
"street": true,
"city": true,
"postcode": true,
"country": false
},
{
"street": true,
"city": true,
"postcode": true,
"country": false
},
{
"street": true,
"city": true,
"postcode": true,
"country": false
},
{
"street": true,
"city": true,
"postcode": true,
"country": false
},
{
"street": true,
"city": true,
"postcode": true,
"country": false
},
{
"street": true,
"city": true,
"postcode": true,
"country": false
},
{
"street": true,
"city": true,
"postcode": false,
"country": false
},
{
"street": true,
"city": true,
"postcode": true,
"country": false
},
{
"street": true,
"city": true,
"postcode": true,
"country": false
}
],
"sample": [
{
"street": "1600 Amphitheatre Parkway",
"unit": null,
"city": "Mountain View",
"region": "CA",
"postcode": "94043",
"country": "USA"
},
{
"street": "10 Downing Street",
"unit": null,
"city": "London",
"region": null,
"postcode": "SW1A 2AA",
"country": "GBR"
},
{
"street": "Rua Augusta 274",
"unit": null,
"city": "Lisbon",
"region": null,
"postcode": "1100-053",
"country": "PT"
}
],
"fixture": "receipts.addresses.v1"
} Weights, every dated article, and repo commits on one rail.
Teal is the weights (2025-02-20), amber triangles are 9 dated articles (each links to it), grey ticks are repo commits.
1,846 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.
From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.
Everything else, folded: the full tables, every source.
| Benchmark | Value | Setting / notes | Reported by | Date |
|---|---|---|---|---|
| measured BFCL · Call tools and APIs | 19.62 percent (overall accuracy) | Prompt5 sub-scores
| leaderboard | 2026-10-25 |
| author-reported AGIEval | 42.1 | Gemma 3 PT 4B | model author | 2026-10-07 |
| author-reported ARC Challenge · Chat and follow instructions | 56.2 | Gemma 3 PT 4B | model author | 2026-10-07 |
| author-reported BBH · Reason through hard problems | 50.9 | Gemma 3 PT 4B | model author | 2026-10-07 |
| author-reported Commonsense QA sets · Chat and follow instructions | 72.3 | Gemma 3 PT 4B | model author | 2026-10-07 |
| author-reported Commonsense QA sets · Chat and follow instructions | 79.6 | Gemma 3 PT 4B | model author | 2026-10-07 |
| author-reported Commonsense QA sets · Chat and follow instructions | 51.9 | Gemma 3 PT 4B | model author | 2026-10-07 |
| author-reported Commonsense QA sets · Chat and follow instructions | 65.8 | Gemma 3 PT 4B | model author | 2026-10-07 |
| author-reported Commonsense QA sets · Chat and follow instructions | 20 | Gemma 3 PT 4B | model author | 2026-10-07 |
| author-reported Commonsense QA sets · Chat and follow instructions | 82.4 | Gemma 3 PT 4B | model author | 2026-10-07 |
| author-reported DROP | 60.1 | Gemma 3 PT 4B | model author | 2026-10-07 |
| author-reported GSM8K · Reason through hard problems | 38.4 | Gemma 3 PT 4B | model author | 2026-10-07 |
| author-reported HellaSwag · Chat and follow instructions | 77.2 | Gemma 3 PT 4B | model author | 2026-10-07 |
| author-reported MATH-500 · Reason through hard problems | 24.2 | Gemma 3 PT 4B | model author | 2026-10-07 |
| author-reported MMLU · Reason through hard problems | 59.6 | Gemma 3 PT 4B | model author | 2026-10-07 |
| author-reported MMLU-Pro · Reason through hard problems | 29.2 | Gemma 3 PT 4B | model author | 2026-10-07 |
| author-reported Winogrande · Chat and follow instructions | 64.7 | Gemma 3 PT 4B | model author | 2026-10-07 |
Gemma 3 4B IT is an instruction-tuned multimodal model with 4.30B parameters.[1][2] It is released under the Gemma Terms of Use and supports a 128K context window.[3][4]
Generated 2026-10-07 14:59 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.
Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.
Listed in LLMs
| Pipeline tag | image-text-to-text |
|---|---|
| Library | transformers |
| Architectures | Gemma3ForConditionalGeneration |
| Model type | gemma3 |
| Dtype | bfloat16 |
| Params by dtype | BF16: 4,300,079,472 |
| Size band | M |
| MoE | no |
| Context length | 131,072 (provider listing) |
| Gated | manual |
| Papers | arXiv:1905.07830, arXiv:1905.10044, arXiv:1911.11641, arXiv:1904.09728, arXiv:1705.03551, arXiv:1911.01547, arXiv:1907.10641, arXiv:1903.00161, arXiv:2009.03300, arXiv:2304.06364, arXiv:2103.03874, arXiv:2110.14168, arXiv:2311.12022, arXiv:2108.07732, arXiv:2107.03374, arXiv:2210.03057, arXiv:2106.03193, arXiv:1910.11856, arXiv:2502.12404, arXiv:2502.21228, arXiv:2404.16816, arXiv:2104.12756, arXiv:2311.16502, arXiv:2203.10244, arXiv:2404.12390, arXiv:1810.12440, arXiv:1908.02660, arXiv:2312.11805 |
| License | Gemma Terms of Use (gemma) |
| Hub snapshot | 2026-10-25 |
vLLM: supported · Gemma3ForConditionalGeneration · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Gemma3ForConditionalGeneration · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed
vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.
Weights-only memory for gemma-3-4b-it: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 10 GB · Single GPU (<24 GB) |
|---|---|
| FP8 (1 byte/param) | 5.2 GB · Single GPU (<24 GB) |
| INT4 (0.5 bytes/param) | 2.6 GB · Edge (<4 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
Base model: google/gemma-3-4b-pt
quantized: 492finetune: 848adapter: 503merge: 3
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| MaziyarPanahi/gemma-3-4b-it-GGUF | quantized | GGUF | — | — | — | MaziyarPanahi | 20 | 3.6M |
| lmstudio-community/gemma-3-4b-it-GGUF | quantized | GGUF | — | — | gemma | lmstudio-community | 31 | 1.8M |
| unsloth/gemma-3-4b-it-unsloth-bnb-4bit | quantized | -bnb | 4.39B | M | gemma | unsloth | 25 | 1.7M |
| unsloth/gemma-3-4b-it | finetune | — | 4.30B | M | gemma | unsloth | 28 | 1.6M |
| unsloth/gemma-3-4b-it-GGUF | quantized | GGUF | — | — | gemma | unsloth | 208 | 1.4M |
| ggml-org/gemma-3-4b-it-GGUF | quantized | GGUF | — | — | gemma | ggml-org | 62 | 666.3k |
| bartowski/google_gemma-3-4b-it-GGUF | quantized | GGUF | — | — | — | bartowski | 39 | 571.9k |
| unsloth/gemma-3-4b-it-bnb-4bit | quantized | -bnb | 4.41B | M | gemma | unsloth | 12 | 434.3k |
| sarvamai/sarvam-translate | finetune | — | 4.30B | M | gpl-3.0 | sarvamai | 163 | 412k |
| farbodtavakkoli/OTel-LLM-4B-IT | finetune | — | 4.30B | M | apache-2.0 | farbodtavakkoli | 1 | 325.6k |
| gaunernst/gemma-3-4b-it-int4-awq | quantized | int4 | 4.30B | M | gemma | gaunernst | 8 | 253.1k |
| soob3123/amoral-gemma3-4B-v1 | quantized | GGUF | 4.30B | M | apache-2.0 | soob3123 | 23 | 190.7k |
| google/gemma-3-4b-it-qat-q4_0-gguf | quantized | q4_0 | — | — | gemma | 313 | 156.4k | |
| lmstudio-community/gemma-3-4B-it-qat-GGUF | quantized | GGUF | — | — | gemma | lmstudio-community | 18 | 152k |
| gaunernst/gemma-3-4b-it-qat-compressed-tensors | quantized | compressed-tensors | 4.40B | M | gemma | gaunernst | 3 | 123k |
| ISTA-DASLab/gemma-3-4b-it-GPTQ-4b-128g | quantized | GPTQ | 4.33B | M | gemma | ISTA-DASLab | 7 | 102.4k |
| RedHatAI/gemma-3-4b-it-quantized.w4a16 | quantized | -quant | 5.00B | M | gemma | RedHatAI | 5 | 91.8k |
| gghfez/gemma-3-4b-novision | finetune | — | 3.88B | M | gemma | gghfez | 7 | 75.7k |
| nvidia/Nemotron-Content-Safety-Reasoning-4B | finetune | — | 4.30B | M | other | nvidia | 35 | 53.2k |
| google/gemma-3-4b-it-qat-int4-unquantized | finetune | int4 | 4.30B | M | gemma | 10 | 47.3k | |
| nvidia/Nemotron-3-Content-Safety | finetune | — | 4.30B | M | other | nvidia | 22 | 45.3k |
| mlabonne/gemma-3-4b-it-abliterated-GGUF | quantized | GGUF | — | — | gemma | mlabonne | 36 | 37.6k |
| RedHatAI/gemma-3-4b-it-FP8-dynamic | quantized | -FP8 | 4.30B | M | apache-2.0 | RedHatAI | 3 | 39.2k |
| DavidAU/gemma-3-4b-it-heretic-uncensored-abliterated-Extreme | finetune | — | 4.30B | M | — | DavidAU | 13 | 37.6k |
| neo4j/text-to-cypher-Gemma-3-4B-Instruct-2025.04.0 | finetune | — | 4.30B | M | gemma | neo4j | 27 | 35.9k |
| google/gemma-3-4b-it-qat-q4_0-unquantized | finetune | q4_0 | 4.30B | M | gemma | 13 | 35.3k | |
| mlabonne/gemma-3-4b-it-abliterated | finetune | — | 4.30B | M | gemma | mlabonne | 31 | 26.1k |
| qiaw99/Gemma3-4b-it-LogiQA-DPO-D | adapter | — | — | — | — | qiaw99 | 0 | 26.7k |
| DavidAU/Gemma-3-4b-it-MAX-NEO-Imatrix-GGUF | quantized | Imatrix | — | — | apache-2.0 | DavidAU | 13 | 24.3k |
| DavidAU/Gemma-3-4b-it-MAX-HORROR-Imatrix-GGUF | quantized | Imatrix | — | — | apache-2.0 | DavidAU | 8 | 23.5k |
| mshojaei77/gemma-3-4b-persian-v0 | quantized | GGUF | 4.30B | M | apache-2.0 | mshojaei77 | 20 | 21.4k |
| aisingapore/Gemma-SEA-LION-v4-4B-VL-GGUF | finetune | GGUF | — | — | gemma | aisingapore | 0 | 21.2k |
| second-state/gemma-3-4b-it-GGUF | quantized | GGUF | — | — | gemma | second-state | 0 | 17.6k |
| INSAIT-Institute/MamayLM-Gemma-3-4B-IT-v1.0 | finetune | — | 4.30B | M | gemma | INSAIT-Institute | 16 | 16.8k |
| Mungert/X-Ray_Alpha-GGUF | quantized | GGUF | — | — | gemma | Mungert | 6 | 17k |
| qiaw99/Gemma3-4b-it-OpenbookQA-DPO-G-v1 | adapter | — | — | — | — | qiaw99 | 0 | 16.8k |
| gaunernst/gemma-3-4b-it-qat-autoawq | quantized | awq | 4.30B | M | gemma | gaunernst | 2 | 16.1k |
| qiaw99/Gemma3-4b-it-OpenbookQA-DPO-F-v1 | adapter | — | — | — | — | qiaw99 | 0 | 16.2k |
| mlabonne/gemma-3-4b-it-abliterated-v2 | finetune | — | 3.88B | M | gemma | mlabonne | 11 | 14.8k |
| qiaw99/Gemma3-4b-it-LogiQA-DPO-F | adapter | — | — | — | — | qiaw99 | 0 | 15.1k |
| Date | Commit | SHA |
|---|---|---|
| 2025-03-21 | Update README (#25) | 093f9f3 |
| 2025-03-21 | Update README.md (#24) | 31fa622 |
| 2025-03-19 | Adding top_k and top_p to generation_config.json | b11f99a |
| 2025-03-12 | Update README.md | dbd91bb |
| 2025-03-12 | Update chat_template (#6) | 34b1dc0 |
| 2025-03-12 | Super-squash branch 'main' using huggingface_hub | cf8f852 |
Hosted endpoints observed for gemma-3-4b-it, sorted by input price (USD per million tokens). Each row names its observer and date; prices change daily, so check the source before quoting. Cheapest observed input price: $0.04/M.
| Provider | Model ref | Context | Quant | $/M in | $/M out | tok/s | TTFT | Tools | Status | Observed |
|---|---|---|---|---|---|---|---|---|---|---|
| Parasail parasail, openrouter | parasail-googlegemma-3-4b-it | 128K | bf16 | $0.04 | $0.05 | — | — | — | serverless | 2026-10-24 |
| DeepInfra hf, deepinfra, openrouter | google/gemma-3-4b-it | 128K | bfloat16 | $0.05 | $0.10 | 90.9 | 327 ms | yes | live | 2026-10-25 |
| Featherless hf | google/gemma-3-4b-it | — | — | — | — | — | — | — | live | 2026-10-25 |
Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.
arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:32 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 14:59 UTC · Last verified 2026-10-25
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.