gpt-oss-20b

20.9B parameters (4 of 32 experts per token), text in, text out, 128K tokens of context. 13 providers serve it from $0.018 per million input tokens.

Verified 2026-10-25 Rank #4 in LLMs · score 136.9 method v1 Sources 5 Downloads 30d 5.8M · #25 of 619 tracked

Changed this week: 3 new quantized builds on the Hub (254 in all).2026-10-24 · 1 new quantized build on the Hub (252 in all).2026-10-24 · 1 new provider: Nscale.2026-10-23

Can I run it?13GB at INT4Estimatefits a 16 GB cardmethodCommunity: We Ran OpenAI GPT-OSS 20B Locally on a Phone →

What does it cost?$0.018/1M incheapest of 10Darkbloomfp82026-10-24

Can I use it?Apache 2.0Permissive; patent grantlicence file2026-10-24

Does it work?5/5probes · 58 tok/sfirst lookparasail2026-10-07

Do people use it?5.8Mdownloads, 30 days#25 of 619 we track · top 4%+23 likes, downloads −4% this week5,214 likesHugging Face2026-10-24

SVG reply: a pelican steals a bicycle from a supermarket — Draw a pelican stealing a bicycle from a supermarket. Reply with one complete SVG document only, no prose and no code fences. Use viewBox="0 0 200 150" and a transparent background. Use only these colours for every fill and stroke: #06e0c0 #ffc298 #cd85f9 #f1f2f2.

Finish the dad joke

low reasoning● 1 line · 10 words

Why did the open-weights model cross the road?

to get to the other side of the loss function

Finish this dad joke. Reply with the punchline only, on one line, without quotes: Why did the open-weights model cross the road?

A parasailer, in ASCII

low reasoning✕ 13×19 over 12×32

      __|__
--o--o--(_)--o--o--
      /   \
     /     \
    /       \
   /         \
  /           \
  ~~~~~~~~~~~~~
  ~~~~~~~~~~~~~
  ~~~~~~~~~~~~~
  ~~~~~~~~~~~~~
  ~~~~~~~~~~~~~
  ~~~~~~~~~~~~~

Draw ASCII art of a parasailer over the sea. At most 12 lines and 32 columns. Reply with the art only: no code fences, no explanation.

Animate a bouncing ball

low reasoning● 8 frames · 16×16

Make a 16x16 pixel sprite animation of a ball bouncing, 8 frames. Reply with JSON only, no code fences: {"palette": {"a": "#rrggbb", ...}, "fps": <number>, "frames": [[16 strings of 16 characters], ...]}. At most 4 palette colours with single-letter keys; "." means transparent; exactly 8 frames.

The ringer: the same four playful tests for every model, replies shown as received and checked by code. A joke, not a benchmark; never ranked.3 of 4 pass · parasail-gpt-oss-20b on parasail · run 2026-10-09 · temperature 0 · cost $0.006 · 2026-10 thing: supermarket · all models →

Worth reading

Picked by rule from 47 tracked pieces: the lab's own post, the most shared, a field report, the latest.

The lab's own post

gpt-oss-120b & gpt-oss-20b Model Card - OpenAI Deployment Safety Hub

deploymentsafety.openai.com 415 words

Most discussed · 490 HN pts

GPT-OSS vs. Qwen3 and a detailed look how things evolved since GPT-2

OpenAI released gpt-oss-120b and gpt-oss-20b, their first open-weight models since GPT-2 in 2019.

A 2025 research paper confirms Dropout results in worse downstream performance in single-epoch regimes.

magazine.sebastianraschka.com 2025-08-10 490 HN pts · 97 comments 5.2k words about the gpt oss family family

Latest · 54 days ago

Show HN: Don't trust me bro: fixing GPT-OSS (3.49B tokens, 1k GPU hours, 1x3090)

burrito-core is a drop-in inference harness for gpt-oss that provides OpenAI and Anthropic API compatibility.

The default jinja template breaks the model, dropping live-test accuracy from ~40% to ~3% when no tools are present.
On vLLM, /v1/responses errors on 73.5–83.5% of multi-turn runs versus 29.0–36.0% on /v1/chat/completions.

github.com 2026-09-01 2 HN pts 1.3k words about the gpt oss family family

25 best of 47 tracked

The 3 picks above, plus these 22. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

ArticleShow HN: Distilling DeepSeek into GPT-OSS doesn't transfer censorship. Try itctgt.ai · 2026-07-30 · 170 HN pts
ArticleRunning GPT-OSS-120B at 500 tokens per second on Nvidia GPUsbaseten.co · 2025-08-07
ArticleIntroducing gpt-ossopenai.com · 2025-08-05 · 175 HN pts · official
ArticleGPT-OSS Reinforcement Learningdocs.unsloth.ai · 2025-09-27 · 180 HN pts
ArticleWhat GPT-OSS leaks about OpenAI's training datafi-le.net · 2025-10-05 · 348 HN pts
DiscussionShow HN: I taught GPT-OSS-120B to see using Google Lens and OpenCVnews.ycombinator.com · 2026-02-11
DiscussionGPT-OSS-120B runs on just 8GB VRAM & 64GB+ system RAMold.reddit.com · 2025-08-11
DiscussionCurious about the training data of OpenAI's new GPT-OSS models? I was tootwitter.com · 2025-08-09 · 239 HN pts
PaperGPT-OSS-20B: A Comprehensive Deployment-Centric Analysis of OpenAI’s Open-Weight Mixture of Experts Modelarxiv.org · undated
PaperIs GPT-OSS Good? A Comprehensive Evaluation of OpenAI’s Latest Open Source Models ∗: Equal contribution. †: Corresponding author: Junhao Song ([email protected])arxiv.org · undated
ArticleIntroducing OpenAI gpt-oss (20b & 120b) - Fireworks AIfireworks.ai · undated
ArticleVisualizing GPT-OSS-20B embeddingsmelonmars.github.io · 2025-08-17
ArticleGPT-OSS 120B Runs at 3000 tokens/sec on Cerebrascerebras.ai · 2025-11-08
ArticleGPT-OSS-Safeguardopenai.com · 2025-10-29 · official
ArticleOpenAI GPT-OSS Model Card [pdf]cdn.openai.com · 2025-08-05 · official
CodeLlama.cpp: Add GPT-OSSgithub.com · 2025-08-05
CodeShow HN: BrowserOS -- browser agents with GPT-OSS, local llmsgithub.com · 2025-08-22
ArticleUsing Codex CLI with GPT-OSS:120B on an Nvidia DGX Spark via Tailscaletil.simonwillison.net · 2025-11-07
ArticleGPT-OSS 20B running on a phonesimonwillison.net · 2025-10-11
CodeShow HN: Run gpt-oss-20b on 8GB GPUsgithub.com · 2025-09-03
CodeOpenAI GPT-OSSgithub.com · 2025-08-05
PaperIs GPT-OSS Good? A Comprehensive Evaluation of OpenAI’s Latest Open Source Modelsarxiv.org · 2025-12-13

Show all 22

Where to run it

Live hosted endpoints, cheapest first, USD per million tokens. Ties break on output price, then provider name.

Prices for the same weights vary 4×

10 providers with a live price, from $0.018 to $0.075 per million input tokens; the median is $0.030. One dot per provider, cheapest in teal.

#	Cheapest 5	In /1M	Out /1M	Context	Quant	Seen
1	Darkbloom	$0.018	$0.090	128K	fp8	Oct 24
2	AkashML	$0.020	$0.10	128K	fp4	Oct 24
3	DeepInfra	$0.024	$0.11	128K	bfloat16	Oct 25
4	DekaLLM	$0.029	$0.14	128K	bf16	Oct 24
5	CoreWeave	$0.030	$0.13	128K	fp4	Oct 24

All 13 providers →

Where it's used

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 (13 API providers); heavily run locally (254 quantized builds); a big base for builders (1,054 fine-tunes, adapters and merges).

As a service13AkashML, Amazon Bedrock, CoreWeave, Darkbloom +9
On own hardware254quantized builds
Built on1,054567 fine-tunes · 480 adapters · 7 merges
In apps100+Hugging Face Spaces

Top Spaces running it: pliny-the-prompter/obliteratus ♥ 621 · umint/ai ♥ 258 · deepset/autoquizzer ♥ 226 · bishmoy/Arxiv-CS-RAG ♥ 195

Will it fit?

Weights only, an estimate: leave room for the KV cache. Method

Fits a 16 GB card at INT4

About 13 GB at INT4, 25 GB at FP8, 50 GB at BF16: weights only, so leave room for the KV cache. Every expert must be resident, so total parameters count.

Estimate params × bytes per param × 1.2 · how we estimate

Day-one receipts

The probe replies as received on 2026-10-07 from parasail, graded by code. Receipts, not a score.

Write a function● pass

def is_balanced(s: str) -> bool:
    stack = []
    pairs = {')': '(', ']': '[', '}': '{'}
    opens = set(pairs.values())
    for ch in s:
        if ch in opens:
            stack.append(ch)
        elif ch in pairs:
            if not stack or stack[-1] != pairs[ch]:
                return False

Needle at 128K● pass

amber-falcon-7731

Tool call● pass

{"city": "Lisbon", "unit": "c"}

Strict JSON● pass

{"name":"Foo-9B-Instruct","license":"Apache-2.0","parameters_b":9.2,"context_tokens":131072,"languages":["English","French","German","Japanese"]}

Speed run● pass

{
  "providers": [
    {
      "provider": "parasail",
      "endpoint_model": "parasail-gpt-oss-20b",
      "source_url": "https://api.parasail.io/v1/chat/completions",
      "ttft_ms_median": 7726,
      "ttft_any_ms_median": 1943,
      "tokens_per_s_median": 77.7,
      "runs": [
        {
          "ttft_s": 8.3,
          "ttft_any_s": 2.128,
          "tokens_per_s": 77.7,
          "latency_s": 10.35,
          "tokens_out": 804,
          "reasoning_chars": 2762,
          "finish": "stop"
        },
        {
          "ttft_s": 2.121,
          "ttft_any_s": 1.525,
          "tokens_per_s": 58.8,
          "latency_s": 4.23,
          "tokens_out": 249,
          "reasoning_chars": 343,
          "finish": "stop"
        },
        {
          "ttft_s": 7.726,
          "ttft_any_s": 1.943,
          "tokens_per_s": 86.6,
          "latency_s": 9.26,
          "tokens_out": 802,
          "reasoning_chars": 2773,
          "finish": "stop"
        }
      ]
    }
  ],
  "note": "Single client, wall clock; tok/s counts every output token including hidden reasoning."
}

Every probe, with timings and cost →

The story so far

Weights, every dated article, and repo commits on one rail.

7 dated articles and 16 repo commits so far

Teal is the weights (2025-08-04), amber triangles are 7 dated articles (each links to it), grey ticks are repo commits.

What people built on it

1,308 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.

1,308 derivatives on the Hub, led by 567 fine-tunes

From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.

GGUF unsloth/gpt-oss-20b-GGUF for llama.cpp, Ollama, LM Studio 5.3M downloads MLX mlx-community/gpt-oss-20b-MXFP4-Q8 for Apple silicon 7.8M downloads FP8 amd/gpt-oss-20b-WFP8-AFP8-KVFP8 for vLLM, SGLang 59.9k downloads

All variants →

Reference

Everything else, folded: the full tables, every source.

What the card says

2 cited sentences

Summary generated, cited

Released under an Apache 2.0 license, this model uses the harmony response format for reasoning and agentic tasks.^[1]^[2] The architecture utilizes 4 experts per token and can run within 16GB of memory using MXFP4 quantization.^[3]^[4]

license
“designed for powerful reasoning, agentic tasks”
experts per token
“the gpt-oss-20b model run within 16GB of memory”

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.

Use-case evidence

6 use cases

What it is good at 6 use cases

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.

Facts

22 fields

Listed in LLMs · MoE

Facts

gpt-oss-20b has 20.9B parameters. (safetensors metadata, observed 2026-10-24)
gpt-oss-20b is a mixture-of-experts model with 32 experts, routing 4 per token. (config.json, observed 2026-10-24)
Context length is 131,072 tokens (128K). (config.json max_position_embeddings, observed 2026-10-24)
License: Apache 2.0 (friction: none). Permissive; patent grant. (license file, observed 2026-10-24)
Created 2025-08-04, last modified 2025-08-26. (HF API, observed 2026-10-24)

Pipeline tag	text-generation
Library	transformers
Architectures	GptOssForCausalLM
Model type	gpt_oss
Params by dtype	BF16: 1,804,459,584 · U8: 19,110,297,600
Size band	L
MoE	yes
Experts	32
Experts per token	4
Context length	131,072 (config.json max_position_embeddings)
Papers	arXiv:2508.10925
License	Apache 2.0 (apache-2.0)
Hub snapshot	2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · GptOssForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · GptOssForCausalLM · 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.

Hardware fit

3 estimates

Hardware fit ESTIMATE

Weights-only memory for gpt-oss-20b: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)	50 GB · Single high-end GPU (<80 GB)
FP8 (1 byte/param)	25 GB · Single high-end GPU (<80 GB)
INT4 (0.5 bytes/param)	13 GB · Single GPU (<24 GB)

Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra. MoE: all experts must be resident, so total parameters count.

First look, every probe

6 probes

First look measured

A fixed probe set run against a live API endpoint: the same questions for every model, answers graded by code, nothing judged by a model. Latencies are single-request wall clock from one client location, not a benchmark.

via parasail parasail-gpt-oss-20b 5/5 probes pass first visible token 2.8 s end-to-end 58 tok/s (median of 3)

Probe	Result	Latency	Reply (truncated)
Self-description	info	6.4 s	I am ChatGPT, a language model based on OpenAI’s GPT‑4 architecture. My training data cutoff is September 2021.
Tool call	pass	2.4 s	{"city": "Lisbon", "unit": "c"}
Strict JSON	pass	4.1 s	{"name":"Foo-9B-Instruct","license":"Apache-2.0","parameters_b":9.2,"context_tokens":131072,"languages":["English","French","German","Japanese"]}
Code (hidden tests)	pass	6.2 s	def is_balanced(s: str) -> bool: stack = [] pairs = {')': '(', ']': '[', '}': '{'} opens = set(pairs.values()) for ch in s: if ch in opens: stack.append(ch) elif ch in pairs: if not stack or stack[-1] != pairs[ch]: return False
Multi-step arithmetic	pass	3.9 s	16
Needle recall at 128k tokens	pass	9.4 s	amber-falcon-7731

Probe set firstlook.v1 · observed 2026-10-07 20:03 UTC · endpoint · provider docs · cost of this run $0.00

All variants

40 variants

Lineage

Original release (shipped quantized by the lab) Family: openai/gpt-oss 1,308 derivatives on the Hub

quantized: 254merge: 7finetune: 567adapter: 480

Top variants (40 shown, ranked by downloads)

Variant	Relation	Quant	Params	Band	License	Author	Likes	Downloads
mlx-community/gpt-oss-20b-MXFP4-Q8	quantized	MXFP4	20.9B	L	apache-2.0	mlx-community	98	7.8M
OpenGVLab/InternVL3_5-GPT-OSS-20B-A4B-Preview-HF	merge	—	21.2B	L	apache-2.0	OpenGVLab	10	6.2M
unsloth/gpt-oss-20b-GGUF	quantized	GGUF	—	—	apache-2.0	unsloth	891	5.3M
unsloth/gpt-oss-20b-BF16	finetune	-BF16	20.9B	L	apache-2.0	unsloth	35	2.8M
lmstudio-community/gpt-oss-20b-GGUF	quantized	GGUF	—	—	apache-2.0	lmstudio-community	89	2.5M
unsloth/gpt-oss-20b-unsloth-bnb-4bit	quantized	-bnb	20.9B	L	apache-2.0	unsloth	41	2.1M
lmstudio-community/gpt-oss-20b-MLX-8bit	quantized	MLX	20.9B	L	apache-2.0	lmstudio-community	60	1.8M
ggml-org/gpt-oss-20b-GGUF	quantized	GGUF	—	—	apache-2.0	ggml-org	184	1.6M
FastFlowLM/GPT-OSS-20B-NPU2	quantized	—	—	—	apache-2.0	FastFlowLM	1	810.2k
OpenGVLab/InternVL3_5-GPT-OSS-20B-A4B-Preview	merge	—	392M	S	apache-2.0	OpenGVLab	86	766.8k
unsloth/gpt-oss-20b	quantized	—	20.9B	L	apache-2.0	unsloth	50	644k
bartowski/openai_gpt-oss-20b-GGUF	quantized	GGUF	—	—	—	bartowski	39	461.8k
RedHatAI/gpt-oss-20b	quantized	—	21.5B	L	apache-2.0	RedHatAI	6	207k
unsloth/gpt-oss-20b-bnb-4bit	quantized	-bnb	20.9B	L	apache-2.0	unsloth	14	200.2k
ArliAI/gpt-oss-20b-Derestricted	finetune	—	20.9B	L	apache-2.0	ArliAI	98	152.8k
MuXodious/gpt-oss-20b-RichardErkhov-heresy	finetune	—	20.9B	L	apache-2.0	MuXodious	27	116.8k
farbodtavakkoli/OTel-LLM-20B-IT	finetune	—	—	—	apache-2.0	farbodtavakkoli	1	105.2k
farbodtavakkoli/OTel-LLM-20B-Reasoning	finetune	—	—	—	apache-2.0	farbodtavakkoli	14	95.5k
Jinx-org/Jinx-gpt-oss-20b	finetune	—	20.9B	L	apache-2.0	Jinx-org	96	58.6k
amd/gpt-oss-20b-MoE-Quant-W-MXFP4-A-FP8-KV-FP8	quantized	-Quant	11.4B	L	apache-2.0	amd	3	59.8k
amd/gpt-oss-20b-WFP8-AFP8-KVFP8	quantized	FP8-	20.9B	L	apache-2.0	amd	0	59.9k
DavidAU/Openai_gpt-oss-20b-NEO-GGUF	quantized	GGUF	—	—	apache-2.0	DavidAU	25	48k
squ11z1/gpt-oss-nano	quantized	GGUF	8.97B	M	apache-2.0	squ11z1	26	42.3k
mlx-community/gpt-oss-20b-MXFP4-Q4	quantized	MXFP4	20.9B	L	apache-2.0	mlx-community	19	40.1k
Mungert/DMind-3-GGUF	quantized	GGUF	—	—	apache-2.0	Mungert	0	37.2k
Mungert/gpt-oss-20b-Derestricted-GGUF	quantized	GGUF	—	—	apache-2.0	Mungert	3	35.6k
chromadb/context-1	finetune	—	20.9B	L	apache-2.0	chromadb	423	10.9k
wangzhang/gpt-oss-20b-abliterated	finetune	—	20.9B	L	apache-2.0	wangzhang	5	31.7k
bartowski/openai_gpt-oss-20b-GGUF-MXFP4-Experimental	quantized	GGUF	—	—	—	bartowski	9	29.5k
mlx-community/gpt-oss-20b-OptiQ-4bit	quantized	-4bit	20.9B	L	apache-2.0	mlx-community	14	28.2k
DavidAU/Openai_gpt-oss-20b-CODER-NEO-CODE-DI-MATRIX-GGUF	quantized	GGUF	—	—	apache-2.0	DavidAU	20	25.1k
DevQuasar/openai.gpt-oss-20b-GGUF	quantized	GGUF	—	—	—	DevQuasar	2	20.6k
InferenceIllusionist/gpt-oss-20b-MLX-4bit	quantized	MLX	20.9B	L	apache-2.0	InferenceIllusionist	20	19.3k
2imi9/gpt-oss-20B-NVFP4A16-BF16	quantized	-BF16	20.9B	L	apache-2.0	2imi9	5	15.7k
KJML/gpt-oss-20b-FP8-Dynamic	quantized	-FP8	20.9B	L	apache-2.0	KJML	1	15.3k
EricSpencer00/chattla-20b	quantized	GGUF	20.9B	L	apache-2.0	EricSpencer00	1	15.1k
manjunathshiva/gpt-oss-20b-tq3	quantized	—	2.94B	M	apache-2.0	manjunathshiva	47	12.4k
giladgd/gpt-oss-20b-GGUF	quantized	GGUF	—	—	apache-2.0	giladgd	2	14.6k
m-polignano/ANITA-NEXT-20B-gpt-oss-ITA	finetune	—	20.9B	L	apache-2.0	m-polignano	5	14.4k
fesalfayed/gpt-oss-20b-hermes_agent-tool-finetune_gguf	quantized	gguf	—	—	apache-2.0	fesalfayed	6	14.3k

Repository changelog

16 commits

changed in the last 7 daysNew providers: 2026-10-23Other: 2026-10-24, 2026-10-24all changes

Repository changelog 16 commits

Date	Commit	SHA
2025-08-26	Update README.md	6cee5e8
2025-08-13	Configure `<\|call\|>` as an EOS token (#109)	d666cf3
2025-08-13	Update README.md	f4770b2
2025-08-08	Update chat_template.jinja (#94)	cbf31f6
2025-08-08	Update README.md	db6852c
2025-08-07	Updated chat template (#82)	10e9d71
2025-08-06	Update the model name in USAGE_POLICY (#52)	2e8f805
2025-08-06	Remove chat_template.json (#59)	ef533c9
2025-08-05	Update README	bbf0930
2025-08-05	Update LICENSE	f47b956
2025-08-05	Update vLLM command	6cd4d0f
2025-08-05	Add model card	21ae1e7
2025-08-05	Upload folder using huggingface_hub	ec854da
2025-08-05	Upload folder using huggingface_hub	5a5e34e
2025-08-04	Upload folder using huggingface_hub	f81fef1
2025-08-04	initial commit	2923e0b

All providers

13 providers · 13 endpoints

Where to run 13 providers · 13 endpoints

Hosted endpoints observed for gpt-oss-20b, 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: $0.02/M.

Provider	Model ref	Context	Quant	$/M in	$/M out	tok/s	TTFT	Tools	Status	Observed
Darkbloom openrouter	openai/gpt-oss-20b	128K	fp8	$0.02	$0.09	—	—	yes	live	2026-10-24
AkashML openrouter	openai/gpt-oss-20b	128K	fp4	$0.02	$0.10	—	—	yes	live	2026-10-24
DeepInfra hf, deepinfra, openrouter	openai/gpt-oss-20b	128K	bfloat16	$0.02	$0.11	127.9	307 ms	yes	live	2026-10-25
DekaLLM openrouter	openai/gpt-oss-20b	128K	bf16	$0.03	$0.14	—	—	yes	live	2026-10-24
Parasail parasail, openrouter	parasail-gpt-oss-20b	128K	fp4	$0.03	$0.15	—	—	yes	serverless	2026-10-24
CoreWeave openrouter	openai/gpt-oss-20b	128K	fp4	$0.03	$0.13	—	—	yes	live	2026-10-24
SiliconFlow openrouter	openai/gpt-oss-20b	128K	fp8	$0.04	$0.18	—	—	—	status -2	2026-10-24
Nscale hf	openai/gpt-oss-20b	128K	—	$0.05	$0.20	171.7	311 ms	yes	live	2026-10-25
OVHcloud hf	gpt-oss-20b	128K	—	$0.05	$0.18	36.6	441 ms	yes	live	2026-10-25
Amazon Bedrock openrouter	openai/gpt-oss-20b	128K	unknown	$0.07	$0.15	—	—	yes	live	2026-10-24
Google openrouter	openai/gpt-oss-20b	128K	unknown	$0.07	$0.25	—	—	—	status -5	2026-10-24
Groq hf, openrouter	openai/gpt-oss-20b	128K	unknown	$0.07	$0.30	721	425 ms	yes	live	2026-10-25
Featherless hf	openai/gpt-oss-20b	—	—	—	—	—	—	—	live	2026-10-25

Articles and coverage

47 pieces

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.

Official3

Introducing gpt-oss

openai.com2025-08-05officialauthorityHN front pageabout the gpt oss familyHN 175 pts · 11 comments

gpt-oss-120b & gpt-oss-20b Model Card - OpenAI Deployment Safety Hub

deploymentsafety.openai.comofficialauthority415 words

GPT-OSS Reinforcement Learning

docs.unsloth.ai2025-09-27authorityHN front pageabout the gpt oss familyHN 180 pts · 43 comments1,473 words

Unsloth enables reinforcement learning training for OpenAI gpt-oss using GRPO, offering 3x faster inference and 50% lower VRAM usage compared to other implementations. Flash Attention 3 causes incorrect training losses due to lack of backward pass support for attention sinks, so Unsloth Flex Attention provides O(N) memory usage instead of O(N^2) for naive attention. Reward hacking countermeasures include inspecting imports, wiping caches, and restricting global variable access to ensure genuine task performance.written, reviewed, cited

Unsloth achieves ~21 tokens/s for gpt-oss RL inference by rewriting code from Transformers since vLLM lacks compatibility.
Training gpt-oss-20b with GRPO is possible on 15GB VRAM using embedding offloading which reduces usage by 1GB.

Analysis6

Show HN: Distilling DeepSeek into GPT-OSS doesn't transfer censorship. Try it

ctgt.ai2026-07-30HN front pageabout the gpt oss familyHN 170 pts · 73 comments2,582 words

A study using DeepSeek V4 Flash as a teacher and GPT-OSS-120B as a student found that political censorship did not transfer during financial reasoning distillation. The LineageEval framework evaluated 304 prompts, revealing the distilled model scored 83.61% on FinanceReasoning at an 8,000-token budget. This approach costs $0.00025939 per query, significantly lower than competing models, while avoiding the teacher's selective avoidance of China-related content.written, reviewed, cited

The distilled 120B model scores 83.61% on FinanceReasoning, outperforming Kimi K3 at 81.93% and Inkling at 65.13%.
The 120B model serves from a single H100 or A100 with base weights in their native MXFP4 at roughly 63 GB.
At an expanded 100,000-token budget, the self-distilled 120B remains stable while Kimi K3 falls to 81.93% and Inkling to 65.13%.

Show HN: Don't trust me bro: fixing GPT-OSS (3.49B tokens, 1k GPU hours, 1x3090)

github.com2026-09-01authorityabout the gpt oss familyHN 2 pts · 0 comments1,271 words

burrito-core is a drop-in inference harness for gpt-oss that provides OpenAI and Anthropic API compatibility. It executes native Python and browser tools within the same process without requiring separate servers, recovering from hallucinated tool calls automatically. A public evaluation suite ran gpt-oss-20b across 320,192 runs consuming 3.49B tokens and 1,062 GPU hours on a single RTX 3090. The project is MIT-licensed and exposes the API on host port 8888.written, reviewed, cited

The default jinja template breaks the model, dropping live-test accuracy from ~40% to ~3% when no tools are present.
On vLLM, /v1/responses errors on 73.5–83.5% of multi-turn runs versus 29.0–36.0% on /v1/chat/completions.
Structured tool schemas triple multi-turn accuracy over AST parsing, where vanilla backends collapse to ~0%.

GPT-OSS vs. Qwen3 and a detailed look how things evolved since GPT-2

magazine.sebastianraschka.com2025-08-10authorityHN front pageabout the gpt oss familyHN 490 pts · 97 comments5,195 words

OpenAI released gpt-oss-120b and gpt-oss-20b, their first open-weight models since GPT-2 in 2019. The 20B model can run on a consumer GPU with up to 16 GB of RAM, and the 120B model can run on a single H100 with 80 GB of RAM or newer hardware. The models utilize MXFP4 optimization to fit onto single GPUs.written, reviewed, cited

A 2025 research paper confirms Dropout results in worse downstream performance in single-epoch regimes.

What GPT-OSS leaks about OpenAI's training data

fi-le.net2025-10-05HN front pageabout the gpt oss familyHN 348 pts · 82 comments3,352 words

GPT-OSS-20B: A Comprehensive Deployment-Centric Analysis of OpenAI’s Open-Weight Mixture of Experts Model

arxiv.orgauthority

Is GPT-OSS Good? A Comprehensive Evaluation of OpenAI’s Latest Open Source Models ∗: Equal contribution. †: Corresponding author: Junhao Song ([email protected])

arxiv.orgauthorityabout the gpt oss family

Community1

Curious about the training data of OpenAI's new GPT-OSS models? I was too

twitter.com2025-08-09HN front pageabout the gpt oss familyHN 239 pts · 57 comments

All coverage tracked for this model (47) — append-only, nothing is ever removed

Title	Source	Published	First seen	Signals
gpt-oss-120b & gpt-oss-20b Model Card - OpenAI Deployment Safety Hub	deploymentsafety.openai.com official	—	2026-10-07	exa
GPT-OSS-20B: A Comprehensive Deployment-Centric Analysis of OpenAI’s Open-Weight Mixture of Experts Model	arxiv.org	—	2026-10-07	exa
Is GPT-OSS Good? A Comprehensive Evaluation of OpenAI’s Latest Open Source Models ∗: Equal contribution. †: Corresponding author: Junhao Song ([email protected])	arxiv.org	—	2026-10-07	exa
Introducing OpenAI gpt-oss (20b & 120b) - Fireworks AI	fireworks.ai	—	2026-10-07	exa
Show HN: Don't trust me bro: fixing GPT-OSS (3.49B tokens, 1k GPU hours, 1x3090)	github.com	2026-09-01	2026-10-07	hn
Show HN: Distilling DeepSeek into GPT-OSS doesn't transfer censorship. Try it	ctgt.ai	2026-07-30	2026-10-07	hn
Real-world GPT-OSS-20B benchmark on L4, L40S and H100 (latency, tokens/SEC)	devforth.io	2026-02-22	2026-10-07	hn
Show HN: I taught GPT-OSS-120B to see using Google Lens and OpenCV	news.ycombinator.com	2026-02-11	2026-10-07	hn
Opencode with Nemotron-3-Nano vs. Qwen3-Coder-30B-A3B vs. GPT-OSS-20B-mxfp4 [video]	youtube.com	2025-12-17	2026-10-07	hn
Is GPT-OSS Good? A Comprehensive Evaluation of OpenAI’s Latest Open Source Models	arxiv.org	2025-12-13	2026-10-07	exa
GPT-OSS 120B Runs at 3000 tokens/sec on Cerebras	cerebras.ai	2025-11-08	2026-10-07	hn
Using Codex CLI with GPT-OSS:120B on an Nvidia DGX Spark via Tailscale	til.simonwillison.net	2025-11-07	2026-10-07	hn
Running LlamaBarn GPT-OSS 20B on my iPhone. Super fast (unedited video)	twitter.com	2025-11-06	2026-10-07	hn
gpt-oss-safeguard: A new milestone for open source safety infrastructure	roost.tools	2025-10-29	2026-10-07	hn
GPT-OSS-Safeguard	openai.com official	2025-10-29	2026-10-07	hn
Compiler optimizations for 5.8ms GPT-OSS-120B inference (not on GPUs)	furiosa.ai	2025-10-17	2026-10-07	hn
We Ran OpenAI GPT-OSS 20B Locally on a Phone	nexa.ai	2025-10-11	2026-10-07	hn
GPT-OSS 20B running on a phone	simonwillison.net	2025-10-11	2026-10-07	hn
What GPT-OSS leaks about OpenAI's training data	fi-le.net	2025-10-05	2026-10-07	hn
GPT-OSS 20B Benchmark A Research Style Synthesis	cognativ.com	2025-10-05	2026-10-07	exa
GPT-OSS Reinforcement Learning	docs.unsloth.ai	2025-09-27	2026-10-07	hn
Winners of OpenAI GPT-OSS-20B Red‑Teaming Challenge	kaggle.com	2025-09-20	2026-10-07	hn
Show HN: Run gpt-oss-20b on 8GB GPUs	github.com	2025-09-03	2026-10-07	hn
Show HN: BrowserOS -- browser agents with GPT-OSS, local llms	github.com	2025-08-22	2026-10-07	hn
GPT-oss from the Ground Up - by Cameron R. Wolfe, Ph.D.	cameronrwolfe.substack.com	2025-08-18	2026-10-07	exa
GPT-OSS-20B at 10k tokens/second on a 4090? Sure	old.reddit.com	2025-08-17	2026-10-07	hn
Visualizing GPT-OSS-20B embeddings	melonmars.github.io	2025-08-17	2026-10-07	hn
GPT-OSS-20B extracted to a base model without alignment	twitter.com	2025-08-14	2026-10-07	hn
GPT-OSS Reasoning and Any-LLM: Nuances of OpenAI API Compatibility	blog.mozilla.ai	2025-08-12	2026-10-07	hn
GPT-OSS-120B runs on just 8GB VRAM & 64GB+ system RAM	old.reddit.com	2025-08-11	2026-10-07	hn
GPT-OSS vs. Qwen3 and a detailed look how things evolved since GPT-2	magazine.sebastianraschka.com	2025-08-10	2026-10-07	hn, exa
OpenAI GPT-OSS models use MXFP4 to cut inference costs	theregister.com	2025-08-10	2026-10-07	hn
Curious about the training data of OpenAI's new GPT-OSS models? I was too	twitter.com	2025-08-09	2026-10-07	hn
Fine-tuning GPT-OSS-20B	colab.research.google.com	2025-08-09	2026-10-07	hn
How to Run Your Own OpenAI GPT OSS Server for Fun and Profit	northcodie.blogspot.com	2025-08-08	2026-10-07	hn
Running GPT-OSS-120B at 500 tokens per second on Nvidia GPUs	baseten.co	2025-08-07	2026-10-07	hn
GPT-OSS 20B delivers GPT-4.1 scores at 5x the speed	dbreunig.com	2025-08-06	2026-10-07	hn
OpenAI's GPT-OSS models benchmarks worse than DeepSeek R1 and Qwen3 235B	xcancel.com	2025-08-06	2026-10-07	hn
GPT-OSS-20B Model on Windows with GPU Acceleration	blogs.windows.com	2025-08-06	2026-10-07	hn
Cerebras now supports OpenAI GPT-OSS-120B at 3k Tokens Per SEC	cerebras.ai	2025-08-06	2026-10-07	hn
Red‑Teaming Challenge – OpenAI GPT-OSS-20B	kaggle.com	2025-08-05	2026-10-07	hn
GPT-OSS is a big deal	twitter.com	2025-08-05	2026-10-07	hn
OpenAI GPT-OSS	github.com	2025-08-05	2026-10-07	hn
OpenAI GPT-OSS Model Card [pdf]	cdn.openai.com official	2025-08-05	2026-10-07	hn
Introducing gpt-oss	openai.com official	2025-08-05	2026-10-07	hn, exa
Llama.cpp: Add GPT-OSS	github.com	2025-08-05	2026-10-07	hn
GPT-OSS 20B \| Awesome Agents	awesomeagents.ai	2025-08-05	2026-10-07	exa

Papers

1 paper

Papers 1

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).

gpt-oss-120b & gpt-oss-20b Model Card this model

OpenAI, Sandhini Agarwal, Lama Ahmad +123 · 2025-08-08 · PDF · HF paper page (30 upvotes)

Similar models

3 models

Similar models

ERNIE-4.5-21B-A3B-Thinking · 21.8B
same size band (L), also MoE, same task (text-generation), permissive license
LFM2.5-24B-A2B · 24.0B
same size band (L), also MoE, same task (text-generation)
GLM-4.7-Flash · 31.2B
same size band (L), also MoE, same task (text-generation), fewer providers, permissive license

Sources

5 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:31 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

Follow gpt-oss-20b

Watch this model

/feeds/models/openai--gpt-oss-20b.xml(RSS 2.0) · every change to gpt-oss-20b
/feeds/families/openai--gpt-oss.xml(RSS 2.0) · every change across the gpt-oss family

Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.

The ringer: four playful tests

A pelican steals a bicycle from a supermarket

Finish the dad joke

A parasailer, in ASCII

Animate a bouncing ball

Worth reading

gpt-oss-120b & gpt-oss-20b Model Card - OpenAI Deployment Safety Hub

GPT-OSS vs. Qwen3 and a detailed look how things evolved since GPT-2

Show HN: Don't trust me bro: fixing GPT-OSS (3.49B tokens, 1k GPU hours, 1x3090)

25 best of 47 tracked

Where to run it

Prices for the same weights vary 4×

Where it's used

Will it fit?

Fits a 16 GB card at INT4

Day-one receipts

The story so far

7 dated articles and 16 repo commits so far

What people built on it

1,308 derivatives on the Hub, led by 567 fine-tunes

Reference

What the card says

Summary generated, cited

Use-case evidence

What it is good at 6 use cases

Facts

Facts

Servability

Servability

Hardware fit

Hardware fit ESTIMATE

First look, every probe

First look measured

All variants

Lineage

Top variants (40 shown, ranked by downloads)

Repository changelog

Repository changelog 16 commits

All providers

Where to run 13 providers · 13 endpoints

Articles and coverage

Official3

Analysis6

Community1

Papers

Papers 1

Similar models

Similar models

Sources

Sources

Follow gpt-oss-20b

Watch this model