SAVRN
Search Contact SAVRN

SAVRN Model Hub · Comparisons

Gemma-4-31B-IT-NVFP4 vs gpt-oss-20b

Gemma-4-31B-IT-NVFP4 has 20.9B parameters and gpt-oss-20b has 20.9B parameters; Gemma-4-31B-IT-NVFP4 is released under other and gpt-oss-20b under Apache License 2.0; at 16-bit, Gemma-4-31B-IT-NVFP4 needs about 50.1 GB (1x MI300X from $1.85 an hour) and gpt-oss-20b about 50.2 GB (1x MI300X from $1.85 an hour).

Published metadata for 2 models, each read from its own repository.
Field Gemma-4-31B-IT-NVFP4
nvidia/Gemma-4-31B-IT-NVFP4
gpt-oss-20b
openai/gpt-oss-20b
Publisher NVIDIA OpenAI
Task Text generation Text generation
Modality Text Text
Parameters, as reported 20.9B parameters 20.9B parameters
Architecture Gemma4ForConditionalGeneration GptOssForCausalLM
Library Model Optimizer transformers
Context length 262,144 tokens 131,072 tokens
Repository size 32.7 GB 41.3 GB
Artifact formats safetensors safetensors
License other apache-2.0
Access Open weights, no gate Open weights, no gate
Memory at 16-bit (weights and margin) 50.1 GB 50.2 GB
Cheapest GPUs at 16-bit, per hour 1x MI300X, $1.85 1x MI300X, $1.85
Memory at 4-bit (weights and margin) 12.5 GB 12.5 GB
Cheapest GPUs at 4-bit, per hour 1x MI300X, $1.85 1x MI300X, $1.85
Revision viewed 4135a98a9b72 6cee5e81ee83
Downloads reported by the hub 1.6M 6.7M
Last observed 2026-09-18 2026-09-18

An evaluation row appears only where at least two of these models report the same benchmark with the same stated configuration, metric, unit and setup. Different evaluators stay named in each cell. Values are shown as reported: no unit conversion, no ranking.

Other Reported Results

These results are listed for each model on its own, because the conditions needed to compare them are not stated or do not match. Two results that leave a condition blank are not assumed to share it.

gpt-oss-20b

BenchmarkConditionsResultReported byRevisionDate
Idavidrein/gpqa Task diamondMetric diamondSetup GPQA DiamondComparison conditions not established 58.5859 EvalEval
Reported by a third party
Evaluated revision not stated 2026-04-19
Idavidrein/gpqa Task diamondMetric diamondSetup Reasoning: mediumComparison conditions not established 66 GPT-OSS Model Card
Reported by a third party
Evaluated revision not stated 2025-08-05
Idavidrein/gpqa Task diamondMetric diamondSetup Reasoning: high, With toolsComparison conditions not established 74.2 GPT-OSS Model Card
Reported by a third party
Evaluated revision not stated 2025-08-05
Idavidrein/gpqa Task diamondMetric diamondSetup Reasoning: highComparison conditions not established 71.5 GPT-OSS Model Card
Reported by a third party
Evaluated revision not stated 2025-08-05
Idavidrein/gpqa Task diamondMetric diamondSetup Reasoning: low, With toolsComparison conditions not established 58 GPT-OSS Model Card
Reported by a third party
Evaluated revision not stated 2025-08-05
Idavidrein/gpqa Task diamondMetric diamondSetup Reasoning: medium, With toolsComparison conditions not established 67.1 GPT-OSS Model Card
Reported by a third party
Evaluated revision not stated 2025-08-05
Idavidrein/gpqa Task diamondMetric diamondSetup Reasoning: lowComparison conditions not established 56.8 GPT-OSS Model Card
Reported by a third party
Evaluated revision not stated 2025-08-05
LEXam-Benchmark/LEXam Task mcq_4_choicesMetric mcq_4_choicesComparison conditions not established 40.78 LEXam Leaderboard
Reported by a third party
Evaluated revision not stated 2026-06-02
LEXam-Benchmark/LEXam Task open_questionMetric open_questionComparison conditions not established 32.12 LEXam Leaderboard
Reported by a third party
Evaluated revision not stated 2026-06-02
LiquidAI/ifstruct-v1.0 Task ifstruct_v1Metric ifstruct_v1Comparison conditions not established 91.95 Liquid AI — IFStruct v1.0 blog (gpt-oss-20b)
Reported by a third party
Evaluated revision not stated 2026-06-30
SWE-bench/SWE-bench_Verified Task swe_bench_%_resolvedMetric swe_bench_%_resolvedSetup Reasoning: lowComparison conditions not established 37.4 GPT-OSS Model Card
Reported by a third party
Evaluated revision not stated 2025-08-05
SWE-bench/SWE-bench_Verified Task swe_bench_%_resolvedMetric swe_bench_%_resolvedSetup Reasoning: mediumComparison conditions not established 53.2 GPT-OSS Model Card
Reported by a third party
Evaluated revision not stated 2025-08-05
SWE-bench/SWE-bench_Verified Task swe_bench_%_resolvedMetric swe_bench_%_resolvedSetup Reasoning: highComparison conditions not established 60.7 GPT-OSS Model Card
Reported by a third party
Evaluated revision not stated 2025-08-05
TIGER-Lab/MMLU-Pro Task mmlu_proMetric mmlu_proComparison conditions not established 73.6 EvalEval
Reported by a third party
Evaluated revision not stated 2026-06-30
cais/hle Task hleMetric hleSetup Reasoning: lowComparison conditions not established 4.2 GPT-OSS Model Card
Reported by a third party
Evaluated revision not stated 2025-08-05
cais/hle Task hleMetric hleSetup Reasoning: medium, With toolsComparison conditions not established 8.8 GPT-OSS Model Card
Reported by a third party
Evaluated revision not stated 2025-08-05
cais/hle Task hleMetric hleSetup Reasoning: mediumComparison conditions not established 7 GPT-OSS Model Card
Reported by a third party
Evaluated revision not stated 2025-08-05
cais/hle Task hleMetric hleSetup Reasoning: highComparison conditions not established 10.9 GPT-OSS Model Card
Reported by a third party
Evaluated revision not stated 2025-08-05
cais/hle Task hleMetric hleSetup Reasoning: high, With toolsComparison conditions not established 17.3 GPT-OSS Model Card
Reported by a third party
Evaluated revision not stated 2025-08-05
cais/hle Task hleMetric hleSetup Reasoning: low, With toolsComparison conditions not established 6.3 GPT-OSS Model Card
Reported by a third party
Evaluated revision not stated 2025-08-05

SAVRN's Notes on Gemma-4-31B-IT-NVFP4

Two numbers on this page pull against each other. The context window is 262,144 tokens, while the 4-bit row, the one that fits an NVFP4 quantization of google/gemma-4-31B-it, needs 12.5 GB of memory for 10.4 GB of weights. On the 192 GB MI300X our Index lists as the cheapest host at $1.85 per hour, the weights are a small tenant; the key-value cache for long inputs is what fills the card, so we size this one by cache, not by weights. The publisher lists text and image input and over 140 languages.

The license is recorded as other, with no commercial-use line on file, so legal reads the terms attached to the gemma-4-31B-it lineage before anything ships. No reported evaluations and no Index host prices exist for this quantization, so a buyer runs their own tests and weighs the 16-bit row's 50.1 GB against the 12.5 GB saved.

SAVRN's Notes on gpt-oss-20b

Five hosts on the SAVRN Index serve gpt-oss-20b by the token, from $0.03 in and $0.14 out per million at DeepInfra to $0.10 and $0.50 at Groq. Owned, this text generator fits on one card: 12.5 GB at 4-bit, 25.1 GB at 8-bit, 50.2 GB at 16-bit, and the least expensive card listed, a 192 GB MI300X, rents for $1.85 an hour, so even full precision leaves room for its 131,072-token context.

You can modify and resell this one; Apache 2.0 allows commercial use, modification and redistribution as long as the notices stay intact. The publisher says it was trained on its harmony response format and should only be used with it, so your serving stack has to speak harmony. And settle rent-or-own yourself: whether $0.14 per million output tokens beats $1.85 an hour depends on how many tokens a day you push through.

Questions

Which is larger, Gemma-4-31B-IT-NVFP4 or gpt-oss-20b?

gpt-oss-20b (20.9B parameters) is larger than Gemma-4-31B-IT-NVFP4 (20.9B parameters), by the parameter counts their publishers report.

Which is cheaper to run, Gemma-4-31B-IT-NVFP4 or gpt-oss-20b?

At 4-bit, Gemma-4-31B-IT-NVFP4 fits on 1x MI300X from $1.85 an hour and gpt-oss-20b on 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.

Can I use gpt-oss-20b commercially?

Yes. gpt-oss-20b is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors.

Related Comparisons