gpt-oss-20b

Name: gpt-oss-20b
Author: OpenAI

Released

2025-08-05

Last refreshed

2026-10-13

Status

Researched 193d ago

Open sourceCommercial use: permittedRAGAgentsLong contextClassificationJSON / Tool use

OpenAI releases · 46 in the last 12 monthsChangelog →

Specifications

Family: GPT-OSS
Released: 2025-08-05
Context: 128K
Parameters: 20B
Architecture: Decoder Only
Knowledge cutoff: 2025-08
Specialization: general
Openness: Open source
License: Apache 2.0OSI-approvedPermitted
Weights: Unknown
Code: Unknown
Training: Pretrained

Created by

OpenAI

Cutting-edge research and development.

San Francisco, California, United States

Founded 2015

Website

Pricing

Output / 1M

Free

Input / 1M

Free

Cheapest of 13 routes · Open Inference

Providers(13)

Cloudflare Workers AI OpenRouter Fireworks AI GCP Vertex AI NVIDIA NIM GroqCloud Replicate API Vercel AI Gateway Novita AI Open Inference OpenAI API Featherless Yandex AI Studio

View 13 provider routes

About

OpenAI open-weight model with 20 billion parameters. Lightweight, efficient text-only model with reasoning and function calling. Free for self-hosting. Released August 5, 2025.

Providers(13)

Compare all 13

Compare API pricing across 13 tracked provider routes for input and output tokens, batch, and cached reads when available.

Provider	Input / 1M	Output / 1M	Cache	Route
Open Inference	Free	Free	-	Serverless
OpenRouter	$0.030	$0.140	-	Serverless
Featherless	$0.040	$0.150	-	Serverless
Novita AI	$0.040	$0.150	-	Serverless
Vercel AI Gateway	$0.050	$0.200	-	Serverless
GCP Vertex AI	$0.070	$0.250	-	Serverless
Fireworks AI	$0.070	$0.300	-	Serverless
GroqCloud	$0.075	$0.300	-	Serverless
Cloudflare Workers AI	$0.200	$0.300	-	Serverless
Replicate API	$0.090	$0.360	-	Serverless
Yandex AI Studio	$0.8197	$0.8197	read $0.8197	Serverless Scheduled retirement: October 30, 2026
OpenAI API	-	-	-	ServerlessPartial
NVIDIA NIM	-	-	-	ServerlessPartial

Available via routers & gateways(15)

LiteLLM

Gateway

Open-source Python SDK and proxy server that unifies 100+ LLM APIs behind a single OpenAI-compatible interface, with load balancing, cost tracking, and configurable failover.

Free OSSOpenAI APIGCP Vertex AI

OpenRouter

Hybrid

Unified hybrid gateway to 400+ models from 60+ providers via a single OpenAI-compatible API, with optional auto-routing that selects the best model per prompt.

PassthroughOpenAI APIGCP Vertex AIFireworks AI

Portkey

Gateway

Production AI gateway routing to 1,600+ LLMs with failover, load balancing, semantic caching, and guardrails; Apache 2.0 core is fully self-hostable with the complete feature set.

SubscriptionOpenAI APIGCP Vertex AI

AIRouter

Router

Commercial LLM router that analyzes incoming requests and routes to the optimal model for cost/quality/latency via a drop-in OpenAI-compatible API, with a privacy-preserving embedding mode that avoids sending prompt content.

Passthrough + feeOpenAI APIGCP Vertex AI

Helicone

Gateway

Observability-first AI gateway with routing, caching, rate limiting, and request tracing; Apache 2.0 open-source core with a managed hosted tier for logging and analytics.

SubscriptionOpenAI APIGCP Vertex AI

Kong AI Gateway

Gateway

Multi-LLM AI gateway built on Kong Gateway 3.x, adding semantic routing, load balancing, guardrails, and MCP traffic analytics as plugins over Kong's existing API management platform.

SubscriptionOpenAI APIGCP Vertex AI

Capabilities

JSON / Tool useStructured Outputs

Benchmark peer barsfor RAG

No task-mapped benchmark peers are available for this model yet.

Benchmark scores(3)

Scores are benchmark-specific and are direction-aware: the same numeric gap can mean very different outcomes across suites. Use the leaderboard context and this model's provider route to decide whether the winning margin is meaningful for your workload.

Benchmark	Score	Version	Evaluation	Source
Chatbot Arena	1317.0	text_style_control overall 2026-10-22 (style control, default view)Observed 2026-10-26	Configuration: gpt-oss-20b Harness: LMArena blind pairwise human votes, Bradley-Terry Evaluator: LMArena Confidence: confirmed Notes: LMArena text style control leaderboard, overall category, published 2026-10-22 (HF lmarena-ai/leaderboard-dataset @9761a3ce, latest split). Board entry 'gpt-oss-20b': rank 269, Arena Score 1317.4 (95% CI 1311.0-1323.8), 10417 votes. Retrieved 2026-10-26.	Source Source: LMArena (lmarena.ai), CC BY 4.0
LMArena Text Arena - Creative Writing	1201.0	text creative_writing 2026-10-22Observed 2026-10-22	Configuration: gpt-oss-20b Harness: LMArena blind pairwise human votes, Bradley-Terry Evaluator: LMArena Confidence: confirmed Notes: LMArena Text arena, Creative Writing category (no style control), board published 2026-10-22 (HF lmarena-ai/leaderboard-dataset @a467e6b9, text config, latest split). Board entry 'gpt-oss-20b' (category creative_writing in text/latest-00000-of-00001.parquet): rank 307 of 416, rating 1201.0 (95% CI 1183.2-1218.7), 1234 votes. Source: LMArena (CC BY 4.0). Retrieved 2026-10-27.	Source Source: LMArena (lmarena.ai), CC BY 4.0
WMT26 General MT (human evaluation, cESA Avg. Rank)	39.8	WMT26 General MT findings, Figure 3 Avg. Rank (lower is better)Observed 2026-10-27	Configuration: GPT OSS 20B Harness: Contrastive Error Span Annotation (cESA), human annotators Evaluator: WMT26 organizers Confidence: confirmed Notes: WMT26 General MT findings (2026.wmt-1.48), Figure 3: Avg. Rank 39.8 for the system listed as 'GPT OSS 20B', averaged over the 23 language pairs it entered; rank 1 is best, so lower is better (not a 0-100 score). Per-pair cESA scores (0-100, higher is better, not comparable across pairs): cs-vi 60.6, cs-de 45.4, en-cs 37.7, en-de 37.5, en-is 24.1, en-zh-Hant 53.3, en-arz 69.9, en-ru 49.5, en-be 21.6, en-lld 1.1, en-se 0.8, en-et 23.3, en-ja 72.4, en-kk 40.8, en-id 56.0, zh-ja 55.1, en-ko 52.0, en-th 31.3, en-lij 2.4, cs-uk 49.7, en-uk 44.6, en-hy 29.0, en-zh-Hans 51.7. Source: WMT26 organizers. Retrieved 2026-10-27.	Source Source: Findings of the WMT26 General Machine Translation Shared Task (Kocmi et al., 2026)