GPT OSS 20B

gpt-oss-20b is an open-weight 21B parameter model released by OpenAI under the Apache 2.0 license. It uses a Mixture-of-Experts (MoE) architecture with 3.6B active parameters per forward pass, optimized for lower-latency inference and deployability on consumer or single-GPU hardware. The model is trained in OpenAI’s Harmony response format and supports reasoning level configuration, fine-tuning, and agentic capabilities including function calling, tool use, and structured outputs.

openai/gpt-oss-20b

Organization: OpenAI
Family: gpt-oss
Providers: 41
Context: 131,072
Output limit: 32,768
Knowledge: 2024-06
Release: 2025-08-05
Updated: —
Weights: Open
Input: text
Output: text
Capabilities: —

Providers

Compare every cataloged offer for this model, including its provider, route, token limits, price, and capabilities.

63 offers · USD per million tokens

Provider	Model ID	Context	Output	Input / output · 1M	Reasoning	Tools	Structured	Details
AI/ML API	openai/gpt-oss-20b	131,000	117,964	— / —	—	—	—	View
Amazon Bedrock	openai.gpt-oss-20b-1:0	131,072	128,000	— / —	—	—	—	View
Amazon Bedrock	openai.gpt-oss-20b	131,072	131,072	— / —	—	—	—	View
Amazon Bedrock	us-gov.openai.gpt-oss-20b-1:0	128,000	16,384	— / —	—	—	—	View
Clarifai	openai/chat-completion/models/gpt-oss-20b	131,072	16,384	— / —	—	—	—	View
Cloudflare Workers AI	@cf/openai/gpt-oss-20b	128,000	16,384	— / —	—	—	—	View
CoreWeave	openai/gpt-oss-20b	131,072	131,072	— / —	—	—	—	View
Cortecs	gpt-oss-20b	131,000	131,000	— / —	—	—	—	View
Databricks	databricks-gpt-oss-20b	131,072	25,000	— / —	—	—	—	View
Deep Infra	openai/gpt-oss-20b	131,072	16,384	— / —	—	—	—	View
DevPass (LLM Gateway)	gpt-oss-20b	131,072	32,766	— / —	—	—	—	View
DigitalOcean	openai-gpt-oss-20b	128,000	4,096	— / —	—	—	—	View
Eden AI	deepinfra/openai/gpt-oss-20b	131,072	32,768	— / —	—	—	—	View
Eden AI	arkanecloud/openai/gpt-oss-20b	131,072	32,768	— / —	—	—	—	View
Eden AI	cloudflare/@cf/openai/gpt-oss-20b	128,000	32,768	— / —	—	—	—	View
Eden AI	groq/openai/gpt-oss-20b	131,072	32,768	— / —	—	—	—	View
Eden AI	databricks/databricks-gpt-oss-20b@eu	131,072	32,768	— / —	—	—	—	View
Eden AI	flexai/gpt-oss-20b	131,072	32,768	— / —	—	—	—	View
Eden AI	databricks/databricks-gpt-oss-20b	131,072	32,768	— / —	—	—	—	View
Eden AI	ovhcloud/gpt-oss-20b	131,072	32,768	— / —	—	—	—	View
Eden AI	greenference/gpt-oss-20b	131,072	32,768	— / —	—	—	—	View
FastRouter	openai/gpt-oss-20b	131,072	32,768	— / —	—	—	—	View
FrogBot	gpt-oss-20b	131,072	32,768	— / —	—	—	—	View
Groq	openai/gpt-oss-20b	131,072	65,536	— / —	—	—	—	View
Helicone	gpt-oss-20b	131,072	131,072	— / —	—	—	—	View
Hugging Face	openai/gpt-oss-20b	131,072	32,768	— / —	—	—	—	View
Impossibl	groq/gpt-oss-20b	131,072	32,768	— / —	—	—	—	View
Impossibl	fireworks/gpt-oss-20b	131,072	32,768	— / —	—	—	—	View
IO.NET	openai/gpt-oss-20b	64,000	4,096	— / —	—	—	—	View
Kenari	gpt-oss-20b	131,072	32,768	— / —	—	—	—	View
Kilo Gateway	openai/gpt-oss-20b	131,072	32,768	— / —	—	—	—	View
LLM Gateway	groq/gpt-oss-20b	131,072	32,766	— / —	—	—	—	View
LLM Gateway	consensusprotocol/gpt-oss-20b	128,000	32,768	— / —	—	—	—	View
LMStudio	openai/gpt-oss-20b	131,072	32,768	— / —	—	—	—	View
LowRouter	auto/openai/gpt-oss-20b	131,072	32,768	— / —	—	—	—	View
Merge Gateway	openai/gpt-oss-20b	128,000	32,000	— / —	—	—	—	View
NanoGPT	openai/gpt-oss-20b	131,072	16,384	— / —	—	—	—	View
Neon	gpt-oss-20b	131,072	25,000	— / —	—	—	—	View
Nvidia	openai/gpt-oss-20b	131,072	32,768	— / —	—	—	—	View
OCI Generative AI	openai.gpt-oss-20b	128,000	16,384	— / —	—	—	—	View
Ollama Cloud	gpt-oss:20b	131,072	32,768	— / —	—	—	—	View
AkashMLvia OpenRouter	openai/gpt-oss-20b fp4	131,072	117,964	$0.02 / $0.10	Yes	Yes	Yes	View
Amazon Bedrockvia OpenRouter	openai/gpt-oss-20b unknown	131,072	117,964	$0.07 / $0.15	Yes	Yes	No	View
Amazon Bedrockvia OpenRouter	openai/gpt-oss-20b unknown	131,072	117,964	$0.07 / $0.15	Yes	Yes	No	View
CoreWeavevia OpenRouter	openai/gpt-oss-20b fp4	131,072	117,964	$0.03 / $0.13	Yes	Yes	Yes	View
Darkbloomvia OpenRouter	openai/gpt-oss-20b fp8	131,072	32,768	$0.018 / $0.09	Yes	Yes	Yes	View
DeepInfravia OpenRouter	openai/gpt-oss-20b bf16	131,072	117,964	$0.03 / $0.14	Yes	Yes	Yes	View
DekaLLMvia OpenRouter	openai/gpt-oss-20b bf16	131,072	117,964	$0.029 / $0.14	Yes	Yes	Yes	View
Googlevia OpenRouter	openai/gpt-oss-20b unknown	131,072	32,768	$0.07 / $0.25	Yes	No	Yes	View
Groqvia OpenRouter	openai/gpt-oss-20b unknown	131,072	65,536	$0.075 / $0.30	Yes	Yes	Yes	View
Parasailvia OpenRouter	openai/gpt-oss-20b fp4	131,072	117,964	$0.03 / $0.15	Yes	Yes	Yes	View
SiliconFlowvia OpenRouter	openai/gpt-oss-20b fp8	131,072	8,192	$0.04 / $0.18	Yes	No	Yes	View
Opper	gpt-oss-20b	128,000	8,192	— / —	—	—	—	View
OVHcloud AI Endpoints	gpt-oss-20b	131,072	131,072	— / —	—	—	—	View
Prime Inference	openai/gpt-oss-20b	131,072	32,768	— / —	—	—	—	View
Qiniu	gpt-oss-20b	128,000	4,096	— / —	—	—	—	View
QVAC	gpt-oss-20b	131,072	32,768	— / —	—	—	—	View
Regolo AI	gpt-oss-20b	128,000	16,384	— / —	—	—	—	View
SiliconFlow	openai/gpt-oss-20b	131,000	8,000	— / —	—	—	—	View
STACKIT	openai/gpt-oss-20b	131,072	8,192	— / —	—	—	—	View
Tempr Gateway	groq/openai/gpt-oss-20b	131,072	65,536	— / —	—	—	—	View
Thinking Machines	openai/gpt-oss-20b	32,768	32,768	— / —	—	—	—	View
Vercel AI Gateway	openai/gpt-oss-20b	131,072	8,192	— / —	—	—	—	View

Prices and limits apply to each offer. A dash means the value is not provided. OpenRouter offers show the hosting provider and routing channel separately.

Pricing

Compare token rates and cache charges across available routes and hosting configurations.

USD per million tokens

Input from$0.018Per 1M input tokens

Output from$0.09Per 1M output tokens

Available offers63Across 41 providers

Lowest listed input and output prices across the offers above; they may belong to different providers. These are base rates, before conditional pricing, discounts or additional fees.

Endpoint price comparisonLatest observed snapshot · 10 lowest-priced configurations

InputOutput

Darkbloomdarkbloom/fp8

$0.018$0.09

AkashMLakashml/fp4

$0.02$0.10

DekaLLMdekallm/bf16

$0.029$0.14

CoreWeavecoreweave/fp4

$0.03$0.13

DeepInfradeepinfra/bf16

$0.03$0.14

Parasailparasail/fp4

$0.03$0.15

SiliconFlowsiliconflow/fp8

$0.04$0.18

Amazon Bedrockamazon-bedrock

$0.07$0.15

Amazon Bedrockamazon-bedrock/eu-west-1

$0.07$0.15

Googlegoogle-vertex/us-central1

$0.07$0.25

openai/gpt-oss-20bObserved Oct 10, 2026, 11:31 AM UTC

Input: $0.018
Output: $0.09
Cache read: $0.009

Hosting prices through OpenRouter All endpoints and cache rates

Endpoint	Input / 1M	Output / 1M	Cache read / 1M	Cache write / 1M
AkashMLakashml/fp4	$0.02	$0.10	—	—
Amazon Bedrockamazon-bedrock	$0.07	$0.15	—	—
Amazon Bedrockamazon-bedrock/eu-west-1	$0.07	$0.15	—	—
CoreWeavecoreweave/fp4	$0.03	$0.13	$0.03	—
Darkbloomdarkbloom/fp8	$0.018	$0.09	$0.009	—
DeepInfradeepinfra/bf16	$0.03	$0.14	—	—
DekaLLMdekallm/bf16	$0.029	$0.14	$0.029	—
Googlegoogle-vertex/us-central1	$0.07	$0.25	—	—
Groqgroq	$0.075	$0.30	$0.0375	—
Parasailparasail/fp4	$0.03	$0.15	$0.02	—
SiliconFlowsiliconflow/fp8	$0.04	$0.18	—	—

Each row is a hosting configuration. Open an offer in Providers for its full pricing conditions and additional fees. Endpoint observations may differ from the routing catalog's base price.

Performance

Review reported latency and output-throughput percentiles from the latest observed rolling window.

Latest 30-minute observation

Time to first token · milliseconds

Latency distributionRolling 30-minute snapshot · 6 endpoints

P50P75P90P99

AkashMLakashml/fp4

791–19,708.24 ms

Amazon Bedrockamazon-bedrock

477–2,382.2 ms

CoreWeavecoreweave/fp4

216–2,758.3 ms

Darkbloomdarkbloom/fp8

3,561–24,469.91 ms

DeepInfradeepinfra/bf16

247–3,993.81 ms

DekaLLMdekallm/bf16

547–14,665.32 ms

All latency percentiles 10 endpoints

Endpoint	P50	P75	P90	P99
AkashMLakashml/fp4	791 ms	1,371 ms	4,328.4 ms	19,708.24 ms
Amazon Bedrockamazon-bedrock	477 ms	604 ms	796 ms	2,382.2 ms
CoreWeavecoreweave/fp4	216 ms	311 ms	407.1 ms	2,758.3 ms
Darkbloomdarkbloom/fp8	3,561 ms	7,151.25 ms	12,006 ms	24,469.91 ms
DeepInfradeepinfra/bf16	247 ms	407 ms	707.9 ms	3,993.81 ms
DekaLLMdekallm/bf16	547 ms	1,102 ms	2,719.6 ms	14,665.32 ms
Googlegoogle-vertex/us-central1	344 ms	787.5 ms	1,425.8 ms	3,304.84 ms
Groqgroq	485 ms	958 ms	1,275 ms	1,987.54 ms
Parasailparasail/fp4	1,293 ms	2,084 ms	2,865.9 ms	6,135.45 ms
SiliconFlowsiliconflow/fp8	1,036 ms	1,265 ms	2,095.9 ms	7,653.5 ms

Output throughput · tokens per second

Throughput distributionRolling 30-minute snapshot · 6 endpoints

P50P75P90P99

AkashMLakashml/fp4

41–163.08 t/s

Amazon Bedrockamazon-bedrock

407–1,014.59 t/s

CoreWeavecoreweave/fp4

221–311 t/s

Darkbloomdarkbloom/fp8

26–72 t/s

DeepInfradeepinfra/bf16

119–161 t/s

DekaLLMdekallm/bf16

35–74 t/s

All throughput percentiles 10 endpoints

Endpoint	P50	P75	P90	P99
AkashMLakashml/fp4	41 t/s	70 t/s	91 t/s	163.08 t/s
Amazon Bedrockamazon-bedrock	407 t/s	524 t/s	655 t/s	1,014.59 t/s
CoreWeavecoreweave/fp4	221 t/s	264 t/s	290 t/s	311 t/s
Darkbloomdarkbloom/fp8	26 t/s	38 t/s	50 t/s	72 t/s
DeepInfradeepinfra/bf16	119 t/s	133 t/s	143 t/s	161 t/s
DekaLLMdekallm/bf16	35 t/s	42 t/s	49 t/s	74 t/s
Googlegoogle-vertex/us-central1	238 t/s	301 t/s	352.1 t/s	411.61 t/s
Groqgroq	180 t/s	347 t/s	600 t/s	895 t/s
Parasailparasail/fp4	66 t/s	81 t/s	94 t/s	115 t/s
SiliconFlowsiliconflow/fp8	69 t/s	92 t/s	119 t/s	151.65 t/s

Percentiles describe the observed request distribution. These are snapshots of a rolling window, not a historical time series.

Uptime

Compare successful request rates across the latest 5-minute, 30-minute, and 24-hour observations.

Successful requests by window

Availability by endpointCurrent overlapping windows · 11 endpoints

Endpoint5 minutes30 minutes24 hours

AkashMLakashml/fp4Observed Oct 10, 2026, 11:31 AM UTC

100%

99.79%

Amazon Bedrockamazon-bedrockObserved Oct 10, 2026, 11:31 AM UTC

100%

Amazon Bedrockamazon-bedrock/eu-west-1Observed Oct 10, 2026, 11:31 AM UTC

—

100%

CoreWeavecoreweave/fp4Observed Oct 10, 2026, 11:31 AM UTC

100%

Darkbloomdarkbloom/fp8Observed Oct 10, 2026, 11:31 AM UTC

100%

DeepInfradeepinfra/bf16Observed Oct 10, 2026, 11:31 AM UTC

100%

99.62%

DekaLLMdekallm/bf16Observed Oct 10, 2026, 11:31 AM UTC

100%

99.93%

Googlegoogle-vertex/us-central1Observed Oct 10, 2026, 11:31 AM UTC

100%

98.87%

GroqgroqObserved Oct 10, 2026, 11:31 AM UTC

20%

43.26%

Parasailparasail/fp4Observed Oct 10, 2026, 11:31 AM UTC

100%

99.93%

SiliconFlowsiliconflow/fp8Observed Oct 10, 2026, 11:31 AM UTC

100%

96.67%

89.21%

Reported by OpenRouter at the time of collection; rate-limited requests are excluded. A dash means no measurement was reported. This is not a live availability check.

Benchmarks

Review attributed evaluation results on each source's original scale.

External evaluations

openai/gpt-oss-20bObserved Oct 10, 2026, 11:31 AM UTC

Benchmark profileArtificial Analysis · shared source scale from 0 to 10

Intelligence

Design Arena	Category	Elo	Win rate	Rank
models	dataviz	935	39.7%	118
models	website	857	27.9%	132

Sources: Artificial Analysis and Design Arena, via OpenRouter. Scores retain their original scales and are not OpenAI Suite ratings.

Detailed evaluations

Evaluation scoresAccuracy or item F1 · higher is better

gpqa diamondOpenAI: gpt-oss-20b

64.205%

tau bench verified airlineOpenAI: gpt-oss-20b

53.876%

Evaluation	Configuration	Source	Results	Details
models · dataviz	GPT OSS 20Bopenai/gpt-oss-20b	Design Arena	933 Elo	Details
gpqa diamond	OpenAI: gpt-oss-20bopenai/gpt-oss-20b	OpenRouter	64.205%	Details
tau bench verified airline	OpenAI: gpt-oss-20bopenai/gpt-oss-20b	OpenRouter	53.876%	Details
Composite indices	gpt-oss-20b (Low)openai/gpt-oss-20b	Artificial Analysis	Intelligence 10 · Coding — · Agentic —	Details
models · website	GPT OSS 20Bopenai/gpt-oss-20b	Design Arena	852 Elo	Details

Sources: Artificial Analysis, Design Arena and OpenRouter. Evaluation configurations and scales differ; results are not interchangeable. Updated Oct 24, 2026.

Apps & session costs

Compare observed median session costs for applications using this model, grouped by conversation length.

30-day sample · through Oct 18, 2026

Application	Model route	Turns	Median cost / session	Details
Hermes Agent	openai/gpt-oss-20b	1 turn	$0.000115	Details
Hermes Agent	openai/gpt-oss-20b	2–9 turns	$0.000765	Details
Hermes Agent	openai/gpt-oss-20b	10–49 turns	$0.009267	Details

Source: OpenRouter, as of Oct 20, 2026. Published under CC BY 4.0. These are observed costs per session, not token prices or a forecast for your workload. Apps are compared separately.

Activity

Daily prompt and completion tokens reported for this model's routes in OpenRouter's top 50. Missing days have no published value.

Last 90 complete UTC days

Daily token usage1,199,792,388,561 reported tokens · 30 days with observations

83B0

2026-08-01
20,007,120,278 tokens

2026-08-02
26,190,729,104 tokens

2026-08-03
27,018,485,849 tokens

2026-08-04
25,741,792,725 tokens

2026-08-05
23,722,184,565 tokens

2026-08-07
19,339,262,034 tokens

2026-08-08
17,078,278,590 tokens

2026-08-09
15,830,224,184 tokens

2026-08-10
19,935,560,396 tokens

2026-08-11
24,180,929,218 tokens

2026-08-12
23,832,008,061 tokens

2026-08-16
19,790,140,301 tokens

2026-08-22
18,516,279,783 tokens

2026-08-24
27,153,001,354 tokens

2026-08-25
29,465,089,337 tokens

2026-08-26
55,479,524,701 tokens

2026-08-27
63,771,547,020 tokens

2026-08-28
64,909,103,373 tokens

2026-08-29
67,189,102,806 tokens

2026-08-30
70,046,628,489 tokens

2026-08-31
83,153,659,746 tokens

2026-09-01
70,098,750,528 tokens

2026-09-02
67,630,538,366 tokens

2026-09-03
62,875,649,305 tokens

2026-09-04
46,673,792,787 tokens

2026-09-05
39,545,040,835 tokens

2026-09-06
43,863,747,369 tokens

2026-09-07
45,685,685,197 tokens

2026-09-08
44,771,168,619 tokens

2026-09-09
36,297,363,641 tokens

Jul 26, 2026Oct 23, 2026

Daily observations · 30 days

Date (UTC)	Reported tokens	Reported routes
Sep 9, 2026	36,297,363,641	1
Sep 8, 2026	44,771,168,619	1
Sep 7, 2026	45,685,685,197	1
Sep 6, 2026	43,863,747,369	1
Sep 5, 2026	39,545,040,835	1
Sep 4, 2026	46,673,792,787	1
Sep 3, 2026	62,875,649,305	1
Sep 2, 2026	67,630,538,366	1
Sep 1, 2026	70,098,750,528	1
Aug 31, 2026	83,153,659,746	1
Aug 30, 2026	70,046,628,489	1
Aug 29, 2026	67,189,102,806	1
Aug 28, 2026	64,909,103,373	1
Aug 27, 2026	63,771,547,020	1
Aug 26, 2026	55,479,524,701	1
Aug 25, 2026	29,465,089,337	1
Aug 24, 2026	27,153,001,354	1
Aug 22, 2026	18,516,279,783	1
Aug 16, 2026	19,790,140,301	1
Aug 12, 2026	23,832,008,061	1
Aug 11, 2026	24,180,929,218	1
Aug 10, 2026	19,935,560,396	1
Aug 9, 2026	15,830,224,184	1
Aug 8, 2026	17,078,278,590	1
Aug 7, 2026	19,339,262,034	1
Aug 5, 2026	23,722,184,565	1
Aug 4, 2026	25,741,792,725	1
Aug 3, 2026	27,018,485,849	1
Aug 2, 2026	26,190,729,104	1
Aug 1, 2026	20,007,120,278	1

Source: OpenRouter (openrouter.ai/rankings), as of Oct 24, 2026. CC BY 4.0. Totals include only individually published routes. Tokenizers differ by provider; the aggregated “other” category is never assigned to a model.

Technical details & API

Explore specifications, supported parameters, reasoning controls, and API identifiers for each OpenRouter route.

Current route specifications

openai/gpt-oss-20bObserved Oct 10, 2026, 11:31 AM UTC

Context: 131,072
Maximum output: 32,768
Tokenizer: GPT
Input: text
Output: text
Moderation: No
Knowledge cutoff: 2024-06-30
Added to OpenRouter: 2025-08-05

Base URLhttps://openrouter.ai/api/v1

Model IDopenai/gpt-oss-20b

Version IDopenai/gpt-oss-20b

Weights IDopenai/gpt-oss-20b

Reasoning

Required: Yes
Enabled by default: Not reported
Default effort: medium
Supported efforts: high, medium, low

Supported parameters

frequency_penalty
include_reasoning
logit_bias
logprobs
max_tokens
min_p
presence_penalty
reasoning
reasoning_effort
repetition_penalty
response_format
seed
stop
structured_outputs
temperature
tool_choice
tools
top_k
top_logprobs
top_p

Frequently asked questions

Answers based on the model metadata and serving offers currently in the catalog.

How much does GPT OSS 20B cost?

Listed rates start at $0.018 per million input tokens and $0.09 per million output tokens. Prices vary by provider and configuration; see Pricing for details.

Which providers offer GPT OSS 20B?

There are 63 cataloged offers across 41 providers. Compare model IDs, prices and limits in Providers.

What is the context window?

The cataloged model context is 131,072 tokens. Each serving endpoint may apply a different limit.

Does it support tools and structured output?

Tool calling: Not reported. Structured output: Not reported. Support can vary by endpoint.

Resources

Hugging Face ↗

Catalog history · 4 recorded revisions

Metadata observations since this model was first cataloged. These are not model release versions.

Oct 21, 2026 · 7dc6ddcb8081
Oct 20, 2026 · b4f71970e464
Sep 14, 2026 · 31b4f2cdfeb5
Sep 12, 2026 · c69dfbe20065