Up to date
google Gemma Terms of Use out 20 months ago
27.4B parameters, images and text in, text out, 128K tokens of context. 6 providers serve it from $0.080 per million input tokens.
Verified 2026-10-26 07:27 UTC Rank #39 in LLMs · score 111.1 method v1 Sources 5 Downloads 30d 341.8k · #202 of 619 tracked
Changed this week: 2 new quantized builds on the Hub (148 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.
MetalChat is a Metal-accelerated C++ framework and command line interpreter for inference of Meta Llama, Google Gemma, and Liquid models.
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.
5 providers with a live price, from $0.080 to $0.12 per million input tokens; the median is $0.080. 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.
Widely served (6 API providers); heavily run locally (148 quantized builds); a big base for builders (742 fine-tunes, adapters and merges).
Top Spaces running it: pliny-the-prompter/obliteratus ♥ 621 · hadadrjt/ai ♥ 86 · TIGER-Lab/MEGA-Bench ♥ 51 · yourbench/advanced ♥ 45
Weights only, an estimate: leave room for the KV cache. Method
About 17 GB at INT4, 33 GB at FP8, 66 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.
29.47. 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 · 2026-10-26
Claim-matched tests, graded by code. Receipts, not a score.
{
"providers": [
{
"provider": "parasail",
"endpoint_model": "parasail-gemma3-27b-it",
"source_url": "https://api.parasail.io/v1/chat/completions",
"ttft_ms_median": 613,
"ttft_any_ms_median": 613,
"tokens_per_s_median": 47.8,
"runs": [
{
"ttft_s": 0.613,
"ttft_any_s": 0.613,
"tokens_per_s": 28.4,
"latency_s": 5.92,
"tokens_out": 168,
"reasoning_chars": 0,
"finish": "stop"
},
{
"ttft_s": 0.647,
"ttft_any_s": 0.647,
"tokens_per_s": 66.1,
"latency_s": 2.81,
"tokens_out": 186,
"reasoning_chars": 0,
"finish": "stop"
},
{
"ttft_s": 0.572,
"ttft_any_s": 0.572,
"tokens_per_s": 47.8,
"latency_s": 3.58,
"tokens_out": 171,
"reasoning_chars": 0,
"finish": "stop"
}
]
}
],
"note": "Single client, wall clock; tok/s counts every output token including hidden reasoning."
} {
"fields_matched": 80,
"fields_total": 80,
"cost_per_1000_usd": 0.0309,
"parsed": true,
"per_item": [
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
}
],
"sample": [
{
"street": "1600 Amphitheatre Parkway",
"unit": null,
"city": "Mountain View",
"region": "CA",
"postcode": "94043",
"country": "US"
},
{
"street": "10 Downing Street",
"unit": null,
"city": "London",
"region": null,
"postcode": "SW1A 2AA",
"country": "GB"
},
{
"street": "Rua Augusta 274",
"unit": null,
"city": "Lisboa",
"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 (2025-03-01), amber triangles are 8 dated articles (each links to it), grey ticks are repo commits.
890 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 Aider Polyglot · Write code with agents | 4.9 percent (pass_rate_2) | edit_format=whole | leaderboard | 2025-03-15 |
| measured BFCL · Call tools and APIs | 29.47 percent (overall accuracy) | Prompt5 sub-scores
| leaderboard | 2026-10-26 |
| author-reported AGIEval | 66.2 | Gemma 3 PT 27B | model author | 2026-10-07 |
| author-reported ARC Challenge · Chat and follow instructions | 70.6 | Gemma 3 PT 27B | model author | 2026-10-07 |
| author-reported BBH · Reason through hard problems | 77.7 | Gemma 3 PT 27B | model author | 2026-10-07 |
| author-reported Commonsense QA sets · Chat and follow instructions | 82.4 | Gemma 3 PT 27B | model author | 2026-10-07 |
| author-reported Commonsense QA sets · Chat and follow instructions | 83.3 | Gemma 3 PT 27B | model author | 2026-10-07 |
| author-reported Commonsense QA sets · Chat and follow instructions | 54.9 | Gemma 3 PT 27B | model author | 2026-10-07 |
| author-reported Commonsense QA sets · Chat and follow instructions | 85.5 | Gemma 3 PT 27B | model author | 2026-10-07 |
| author-reported Commonsense QA sets · Chat and follow instructions | 36.1 | Gemma 3 PT 27B | model author | 2026-10-07 |
| author-reported Commonsense QA sets · Chat and follow instructions | 89 | Gemma 3 PT 27B | model author | 2026-10-07 |
| author-reported DROP | 77.2 | Gemma 3 PT 27B | model author | 2026-10-07 |
| author-reported GSM8K · Reason through hard problems | 82.6 | Gemma 3 PT 27B | model author | 2026-10-07 |
| author-reported HellaSwag · Chat and follow instructions | 85.6 | Gemma 3 PT 27B | model author | 2026-10-07 |
| author-reported MATH-500 · Reason through hard problems | 50 | Gemma 3 PT 27B | model author | 2026-10-07 |
| author-reported MMLU · Reason through hard problems | 78.6 | Gemma 3 PT 27B | model author | 2026-10-07 |
| author-reported MMLU-Pro · Reason through hard problems | 52.2 | Gemma 3 PT 27B | model author | 2026-10-07 |
| author-reported Winogrande · Chat and follow instructions | 78.8 | Gemma 3 PT 27B | model author | 2026-10-07 |
The model handles image and text inputs to generate text outputs.[1] It is released under the Gemma Terms of Use and supports a 128K context window.[2][3]
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: 27,432,406,640 |
| Size band | L |
| 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 | 2026-10-26 |
vLLM: supported · Gemma3ForConditionalGeneration · 4b1c04e76fe9ac3ca8b7e541dd423e7d5f38dc60SGLang: supported · Gemma3ForConditionalGeneration · 60d941293e3af27a12c6eb64e2c3e2e1a17f100f
vLLM registry (4b1c04e76fe9ac3ca8b7e541dd423e7d5f38dc60) · SGLang registry (60d941293e3af27a12c6eb64e2c3e2e1a17f100f) · 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-3-27b-it: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 66 GB · Single high-end GPU (<80 GB) |
|---|---|
| FP8 (1 byte/param) | 33 GB · Single high-end GPU (<80 GB) |
| INT4 (0.5 bytes/param) | 17 GB · Single GPU (<24 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
Base model: google/gemma-3-27b-pt
quantized: 148finetune: 454merge: 5adapter: 283
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| gaunernst/gemma-3-27b-it-int4-awq | quantized | int4 | 27.4B | L | gemma | gaunernst | 50 | 5.2M |
| ISTA-DASLab/gemma-3-27b-it-GPTQ-4b-128g | quantized | GPTQ | 27.6B | L | gemma | ISTA-DASLab | 44 | 3.5M |
| MaziyarPanahi/gemma-3-27b-it-GGUF | quantized | GGUF | — | — | — | MaziyarPanahi | 8 | 3M |
| pytorch/gemma-3-27b-it-AWQ-INT4 | quantized | AWQ | — | — | apache-2.0 | pytorch | 7 | 2.3M |
| farbodtavakkoli/OTel-LLM-27B-IT | finetune | — | — | — | apache-2.0 | farbodtavakkoli | 0 | 2.2M |
| RedHatAI/gemma-3-27b-it-quantized.w4a16 | quantized | -quant | 28.8B | L | gemma | RedHatAI | 13 | 2M |
| unsloth/gemma-3-27b-it-GGUF | quantized | GGUF | — | — | gemma | unsloth | 208 | 1.3M |
| RedHatAI/gemma-3-27b-it-FP8-dynamic | quantized | -FP8 | 27.4B | L | apache-2.0 | RedHatAI | 13 | 1.2M |
| mlabonne/gemma-3-27b-it-abliterated | finetune | — | 27.4B | L | gemma | mlabonne | 345 | 733.6k |
| unsloth/gemma-3-27b-it-bnb-4bit | quantized | -bnb | 28.2B | L | gemma | unsloth | 19 | 621.5k |
| unsloth/gemma-3-27b-it | finetune | — | 27.4B | L | gemma | unsloth | 16 | 594.8k |
| lmstudio-community/gemma-3-27B-it-qat-GGUF | quantized | GGUF | — | — | gemma | lmstudio-community | 19 | 549.9k |
| mlabonne/gemma-3-27b-it-abliterated-GGUF | quantized | GGUF | — | — | gemma | mlabonne | 252 | 455.5k |
| lmstudio-community/gemma-3-27b-it-GGUF | quantized | GGUF | — | — | gemma | lmstudio-community | 51 | 444.7k |
| google/gemma-3-27b-it-qat-q4_0-gguf | quantized | q4_0 | — | — | gemma | 407 | 419.2k | |
| m8than/gemma-3-27b-lenientchatfix | finetune | — | 27.4B | L | gemma | m8than | 0 | 298.5k |
| leon-se/gemma-3-27b-it-FP8-Dynamic | quantized | -FP8 | 27.4B | L | gemma | leon-se | 20 | 294.1k |
| bartowski/google_gemma-3-27b-it-GGUF | quantized | GGUF | — | — | gemma | bartowski | 66 | 284.6k |
| unsloth/gemma-3-27b-it-unsloth-bnb-4bit | quantized | -bnb | 28.2B | L | gemma | unsloth | 19 | 281.9k |
| google/gemma-3-27b-it-qat-q4_0-unquantized | finetune | q4_0 | 27.4B | L | gemma | 42 | 267.3k | |
| gaunernst/gemma-3-27b-it-qat-compressed-tensors | quantized | compressed-tensors | 28.2B | L | gemma | gaunernst | 13 | 87.7k |
| ggml-org/gemma-3-27b-it-GGUF | quantized | GGUF | — | — | gemma | ggml-org | 24 | 74.4k |
| gaunernst/gemma-3-27b-it-qat-autoawq | quantized | awq | 27.4B | L | gemma | gaunernst | 12 | 62.4k |
| abhishekchohan/gemma-3-27b-it-quantized-W4A16 | quantized | -quant | 27.6B | L | gemma | abhishekchohan | 5 | 51.2k |
| pytorch/gemma-3-27b-it-FP8 | quantized | -FP8 | — | — | apache-2.0 | pytorch | 3 | 45.4k |
| pytorch/gemma-3-27b-it-INT4 | quantized | INT4 | 5.57B | M | apache-2.0 | pytorch | 0 | 33.6k |
| TheDrummer/Big-Tiger-Gemma-27B-v3-GGUF | quantized | GGUF | — | — | — | TheDrummer | 27 | 27.4k |
| RedHatAI/gemma-3-27b-it-quantized.w8a8 | quantized | -quant | 28.8B | L | gemma | RedHatAI | 7 | 28.3k |
| mradermacher/gemma-3-27b-it-GGUF | quantized | GGUF | — | — | gemma | mradermacher | 5 | 20.7k |
| NeoChen1024/gemma-3-27b-it-NVFP4 | quantized | NVFP4 | 17.6B | L | gemma | NeoChen1024 | 3 | 17.5k |
| DevQuasar-10/google.gemma-3-27b-it-GGUF | quantized | GGUF | — | — | — | DevQuasar-10 | 0 | 16.9k |
| NbAiLab/borealis-27b | finetune | — | 27.4B | L | other | NbAiLab | 10 | 15.7k |
| second-state/gemma-3-27b-it-GGUF | quantized | GGUF | — | — | gemma | second-state | 0 | 15.5k |
| neo4j/text-to-cypher-Gemma-3-27B-Instruct-2025.04.0 | finetune | — | 27.4B | L | gemma | neo4j | 11 | 14.2k |
| TheDrummer/Gemma-3-R1-27B-v1-GGUF | quantized | GGUF | — | — | — | TheDrummer | 13 | 13.7k |
| TheDrummer/Big-Tiger-Gemma-27B-v3 | finetune | — | — | — | — | TheDrummer | 61 | 11k |
| TheCluster/gemma-3-27b-it-uncensored-mlx-4bit | quantized | mlx | — | — | gemma | TheCluster | 4 | 13.7k |
| Tesslate/Synthia-S1-27b | finetune | — | 27.4B | L | gemma | Tesslate | 102 | 8.5k |
| mradermacher/gemma-3-27b-it-i1-GGUF | quantized | GGUF | — | — | gemma | mradermacher | 4 | 12.1k |
| MISHANM/google-gemma-3-27b-it-fp8 | quantized | -fp8 | 27.4B | L | — | MISHANM | 3 | 11.7k |
| Date | Commit | SHA |
|---|---|---|
| 2025-03-21 | Update README.md (#35) | 005ad34 |
| 2025-03-21 | Update README.md (#34) | d2a0c03 |
| 2025-03-19 | Adding top_k and top_p to generation_config.json | 1ead709 |
| 2025-03-12 | Update README.md | dfb98f2 |
| 2025-03-12 | Update chat_template (#3) | 8ae719a |
| 2025-03-12 | Super-squash branch 'main' using huggingface_hub | da8b0ba |
Hosted endpoints observed for gemma-3-27b-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: $0.08/M.
| Provider | Model ref | Context | Quant | $/M in | $/M out | tok/s | TTFT | Tools | Status | Observed |
|---|---|---|---|---|---|---|---|---|---|---|
| DeepInfra hf, deepinfra, openrouter | google/gemma-3-27b-it | 128K | fp8 | $0.08 | $0.16 | 33.1 | 310 ms | yes | live | 2026-10-26 |
| Parasail parasail, openrouter | parasail-gemma3-27b-it | 128K | fp8 | $0.08 | $0.45 | — | — | — | serverless | 2026-10-26 |
| NextBit openrouter | google/gemma-3-27b-it | 128K | fp8 | $0.08 | $0.35 | — | — | — | live | 2026-10-26 |
| Nebius openrouter | google/gemma-3-27b-it | 107K | fp8 | $0.10 | $0.30 | — | — | — | live | 2026-10-26 |
| Novita openrouter | google/gemma-3-27b-it | 96K | bf16 | $0.12 | $0.20 | — | — | — | live | 2026-10-26 |
| Featherless hf | google/gemma-3-27b-it | — | — | — | — | — | — | — | live | 2026-10-26 |
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: 2026-10-26 · HF API: 2026-10-26 07:14 UTC · Providers: 2026-10-26 07:26 UTC · Articles: 2026-10-26 07:27 UTC · Summary: 2026-10-07 14:59 UTC · Last verified 2026-10-26 07:27 UTC
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.