gpt-oss-120b

117B parameters (4 of 128 experts per token), text in, text out, 128K tokens of context. 27 providers serve it from $0.030 per million input tokens.

Verified 2026-10-25 Rank #2 in LLMs · score 138.8 method v1 Sources 5 Downloads 30d 2.9M · #50 of 619 tracked

Changed this week: 1 new provider: Nscale.2026-10-23

Can I run it?70GB at INT4Estimatefits 2×80 GBmethodCommunity: Show HN: ExpertCache – Run the Full GPT-OSS 120B (63GB) on a 16GB M1 Pro →

What does it cost?$0.030/1M incheapest of 25DeepInfrabfloat162026-10-25

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

Does it work?5/5probes · 168 tok/sfirst lookparasail2026-10-07

Do people use it?2.9Mdownloads, 30 days#50 of 619 we track · top 8%+22 likes, downloads −13% this week5,438 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

as received● 1 line · 9 words

Why did the open-weights model cross the road?

To get to the other side of the dataset.

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

as received● 9×19 fits 12×32

        __/\\__        
       /      \       
      /  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

as received● 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.4 of 4 pass · parasail-gpt-oss-120b on parasail · run 2026-10-09 · temperature 0 · cost $0.004 · 2026-10 thing: supermarket · all models →

Worth reading

Picked by rule from 44 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 · 3 months ago

Show HN: ExpertCache – Run the Full GPT-OSS 120B (63GB) on a 16GB M1 Pro

The 16 GiB M1 Pro path completed the seven-scenario 8K qualification in 8,249 seconds total.
The successful 16 GiB configuration explicitly disabled automatic fit and whole-file mmap prefetch.

github.com 2026-08-08 4 HN pts 1.3k words

25 best of 44 tracked

The 4 picks above, plus these 21. 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 · 247 HN pts
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 · 43 HN pts
DiscussionCurious about the training data of OpenAI's new GPT-OSS models? I was tootwitter.com · 2025-08-09
PaperIs GPT-OSS Good? A Comprehensive Evaluation of OpenAI's ...arxiv.org · undated
Articlegpt-oss-120b — Architecture, Evals & Inference Performance | InferenceXinferencex.semianalysis.com · 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-120B (high): API Provider Performance Benchmarkingartificialanalysis.ai · 2025-08-13
CodeShow HN: Run gpt-oss-20b on 8GB GPUsgithub.com · 2025-09-03
CodeOpenAI GPT-OSSgithub.com · 2025-08-05
CodeShow HN: We let GPT OSS 120B write and run Python and ARC AGI 2 jumped 4xgithub.com · 2026-02-11

Show all 21

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 12×

25 providers with a live price, from $0.030 to $0.35 per million input tokens; the median is $0.10. One dot per provider, cheapest in teal.

#	Cheapest 5	In /1M	Out /1M	Context	Quant	Seen
1	DeepInfra	$0.030	$0.14	128K	bfloat16	Oct 25
2	Near AI	$0.030	$0.15	128K	fp4	Oct 24
3	Venice	$0.030	$0.15	125K	int4	Oct 24
4	CoreWeave	$0.030	$0.17	128K	fp4	Oct 24
5	DekaLLM	$0.030	$0.18	128K	bf16	Oct 24

All 27 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 (27 API providers); heavily run locally (134 quantized builds); a big base for builders (370 fine-tunes, adapters and merges).

As a service27AkashML, Amazon Bedrock, Baseten, Cerebras +23
On own hardware134quantized builds
Built on370124 fine-tunes · 245 adapters · 1 merges
In apps100+Hugging Face Spaces

Top Spaces running it: hadadxyz/ai ♥ 819 · ginipick/FLUXllama ♥ 561 · smolagents/ml-intern ♥ 423 · burtenshaw/karpathy-llm-council ♥ 292

Will it fit?

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

Fits two 80 GB GPUs at INT4; too big for one

About 70 GB at INT4, 140 GB at FP8, 280 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

How it scores

Third-party measurements only; author-reported numbers stay in Reference, marked as such.

#6 of 13 on Aider Polyglot so far

41.8. Each tick is a model on this site with a measured Aider Polyglot row; author-reported numbers stay in Reference. Percent (pass_rate_2). edit_format=diff

Source · leaderboard · 2025-08-06

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:
    pairs = {')': '(', ']': '[', '}': '{'}
    stack = []
    for ch in s:
        if ch in '([{':
            stack.append(ch)
        elif ch in ')]}':
            if not stack or stack[-1] != pairs[ch]:
                return False
            stack.pop()
    retu

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-120b",
      "source_url": "https://api.parasail.io/v1/chat/completions",
      "ttft_ms_median": 889,
      "ttft_any_ms_median": 589,
      "tokens_per_s_median": 101.5,
      "runs": [
        {
          "ttft_s": 0.889,
          "ttft_any_s": 0.589,
          "tokens_per_s": 129.8,
          "latency_s": 1.86,
          "tokens_out": 241,
          "reasoning_chars": 184,
          "finish": "stop"
        },
        {
          "ttft_s": 1.123,
          "ttft_any_s": 0.592,
          "tokens_per_s": 96.8,
          "latency_s": 2.55,
          "tokens_out": 247,
          "reasoning_chars": 197,
          "finish": "stop"
        },
        {
          "ttft_s": 0.875,
          "ttft_any_s": 0.575,
          "tokens_per_s": 101.5,
          "latency_s": 2.37,
          "tokens_out": 241,
          "reasoning_chars": 172,
          "finish": "stop"
        }
      ]
    }
  ],
  "note": "Single client, wall clock; tok/s counts every output token including hidden reasoning."
}

Three-tool errand● pass

{
  "calls": [
    {
      "name": "get_weather",
      "args": {
        "city": "Lisbon",
        "date": "2026-10-14"
      }
    },
    {
      "name": "check_calendar",
      "args": {
        "date": "2026-10-14"
      }
    },
    {
      "name": "book_table",
      "args": {
        "restaurant": "Cervejaria Ramiro",
        "date": "2026-10-14",
        "time": "19:30",
        "party_size": 2
      }
    }
  ],
  "checks": {
    "get_weather": true,
    "check_calendar": true,
    "book_table": true
  },
  "final": "The reservation at Cervejaria Ramiro for two people on Wednesday, 2026‑10‑14 at 19:30 has been confirmed. Your confirmation code is **TBL-4471**. Enjoy your dinner!",
  "fixture": "receipts.errand.v1"
}

Every probe, with timings and cost →

The story so far

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

9 dated articles and 17 repo commits so far

Teal is the weights (2025-08-04), amber triangles are 9 dated articles (each links to it), grey ticks are repo commits.

What people built on it

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

504 derivatives on the Hub, led by 245 adapters

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

GGUF ggml-org/gpt-oss-120b-GGUF for llama.cpp, Ollama, LM Studio 2.9M downloads MLX lmstudio-community/gpt-oss-120b-MLX-8bit for Apple silicon 1.1M downloads FP8 amd/gpt-oss-120b-w-mxfp4-a-fp8 for vLLM, SGLang 122.7k downloads

All variants →

Reference

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

Benchmarks

2 rows

Benchmarks 2 rows

measured third-party leaderboard row (2)author-reported number from the model card or eval filesNo confidence intervals unless the source gave them; settings differ between rows, so compare within a benchmark, not across. Date = the benchmark's own date when it gives one, else when we read it.

Benchmark	Value	Setting / notes	Reported by	Date
measured Aider Polyglot · Write code with agents	41.8 percent (pass_rate_2)	edit_format=diff	leaderboard	2025-08-06
measured SWE-bench Verified · Write code with agents	26 percent resolved	agent: mini-SWE-agent	leaderboard	2025-08-07

Use-case evidence

5 use cases

What it is good at 5 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-120b has 117B parameters. (safetensors metadata, observed 2026-10-24)
gpt-oss-120b is a mixture-of-experts model with 128 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: 2,167,371,072 · U8: 114,661,785,600
Size band	XXL
MoE	yes
Experts	128
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-120b: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)	280 GB · Multi-GPU node (<640 GB)
FP8 (1 byte/param)	140 GB · Multi-GPU node (<640 GB)
INT4 (0.5 bytes/param)	70 GB · Single high-end GPU (<80 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-120b 5/5 probes pass first visible token 0.8 s end-to-end 168 tok/s (median of 3)

Probe	Result	Latency	Reply (truncated)
Self-description	info	1.3 s	I am ChatGPT, a large language model built on OpenAI’s GPT‑4 architecture and created by OpenAI. My training data includes information up to June 2024.
Tool call	pass	0.7 s	{"city": "Lisbon", "unit": "c"}
Strict JSON	pass	2.3 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	4.5 s	def is_balanced(s: str) -> bool: pairs = {')': '(', ']': '[', '}': '{'} stack = [] for ch in s: if ch in '([{': stack.append(ch) elif ch in ')]}': if not stack or stack[-1] != pairs[ch]: return False stack.pop() retu
Multi-step arithmetic	pass	1.2 s	16
Needle recall at 128k tokens	pass	6.2 s	amber-falcon-7731

Probe set firstlook.v1 · observed 2026-10-07 20:03 UTC · endpoint · provider docs · cost of this run $0.01

All variants

40 variants

Lineage

Original release (shipped quantized by the lab) Family: openai/gpt-oss 504 derivatives on the Hub

quantized: 134finetune: 124adapter: 245merge: 1

Top variants (40 shown, ranked by downloads)

Variant	Relation	Quant	Params	Band	License	Author	Likes	Downloads
ggml-org/gpt-oss-120b-GGUF	quantized	GGUF	—	—	apache-2.0	ggml-org	81	2.9M
unsloth/gpt-oss-120b-GGUF	quantized	GGUF	—	—	apache-2.0	unsloth	298	2.1M
lmstudio-community/gpt-oss-120b-MLX-8bit	quantized	MLX	117B	XXL	apache-2.0	lmstudio-community	15	1.1M
unsloth/gpt-oss-120b	quantized	—	117B	XXL	apache-2.0	unsloth	23	684.7k
unsloth/gpt-oss-120b-BF16	finetune	-BF16	117B	XXL	apache-2.0	unsloth	9	615k
lmstudio-community/gpt-oss-120b-GGUF	quantized	GGUF	—	—	apache-2.0	lmstudio-community	19	404.5k
unsloth/gpt-oss-120b-unsloth-bnb-4bit	quantized	-bnb	117B	XXL	apache-2.0	unsloth	20	294.5k
openai/gpt-oss-safeguard-120b	finetune	—	117B	XXL	apache-2.0	openai	120	238.2k
nvidia/gpt-oss-120b-Eagle3-short-context	finetune	—	787M	S	other	nvidia	22	238.2k
nvidia/gpt-oss-120b-Eagle3-long-context	finetune	—	215M	S	other	nvidia	91	161.3k
mlx-community/gpt-oss-120b-MXFP4-Q8	quantized	MXFP4	117B	XXL	apache-2.0	mlx-community	21	131.7k
amd/gpt-oss-120b-w-mxfp4-a-fp8	quantized	mxfp4	59.5B	XL	apache-2.0	amd	10	122.7k
bartowski/openai_gpt-oss-120b-GGUF	quantized	GGUF	—	—	—	bartowski	13	122.2k
HauhauCS/GPTOSS-120B-Uncensored-HauhauCS-Aggressive	quantized	GGUF	—	—	apache-2.0	HauhauCS	95	107k
nvidia/gpt-oss-120b-Eagle3-v3	finetune	—	787M	S	other	nvidia	16	99.6k
tiny-random/gpt-oss-bf16	finetune	-bf16	6.85M	S	—	tiny-random	1	97.8k
twhitworth/gpt-oss-120b-awq-w4a16	finetune	awq	117B	XXL	apache-2.0	twhitworth	24	81.8k
DavidAU/Openai_gpt-oss-120b-NEO-Imatrix-GGUF	quantized	Imatrix	—	—	apache-2.0	DavidAU	26	59.2k
RedHatAI/gpt-oss-120b	quantized	—	120B	XXL	apache-2.0	RedHatAI	6	51.1k
yujiepan/gpt-oss-tiny-random	finetune	—	6.85M	S	—	yujiepan	2	39.6k
DevQuasar/openai.gpt-oss-120b-GGUF	quantized	GGUF	—	—	—	DevQuasar	2	23.5k
mlx-community/gpt-oss-120b-MXFP4-Q4	quantized	MXFP4	117B	XXL	apache-2.0	mlx-community	7	20.2k
tiny-random/gpt-oss-mxfp4	quantized	mxfp4	6.87M	S	—	tiny-random	0	20.2k
ArliAI/gpt-oss-120b-Derestricted	finetune	—	117B	XXL	apache-2.0	ArliAI	87	15.7k
gabriellarson/gpt-oss-120b-GGUF	quantized	GGUF	—	—	apache-2.0	gabriellarson	5	17.1k
bartowski/openai_gpt-oss-120b-GGUF-MXFP4-Experimental	quantized	GGUF	—	—	—	bartowski	12	16.5k
NexVeridian/gpt-oss-120b-3bit	quantized	—	117B	XXL	apache-2.0	NexVeridian	6	11.5k
nvidia/gpt-oss-120b-Eagle3-throughput	finetune	—	762M	S	other	nvidia	38	9.8k
giladgd/gpt-oss-120b-GGUF	quantized	GGUF	—	—	apache-2.0	giladgd	1	11.4k
FreeSEED-AI/gpt-oss-120b-mandarin-thinking	finetune	—	117B	XXL	apache-2.0	FreeSEED-AI	51	8.2k
cloudyu/GPT-OSS-120B-2experts-MLX-q4-Claude-4.6-Opus-Reasoning-Distilled	adapter	MLX	117B	XXL	other	cloudyu	9	9.7k
ZERO-POINT-AI/undermaintenance__118b-gpt-faulty	quantized	GGUF	117B	XXL	apache-2.0	ZERO-POINT-AI	6	9k
MultiverseComputingCAI/HyperNova-60B	quantized	—	58.7B	XL	apache-2.0	MultiverseComputingCAI	62	5.8k
furiosa-ai/gpt-oss-120b	quantized	—	117B	XXL	apache-2.0	furiosa-ai	0	8.7k
tiny-random/gpt-oss	finetune	—	6.85M	S	—	tiny-random	2	7.8k
yujiepan/gpt-oss-tiny-random-mxfp4	quantized	mxfp4	6.87M	S	—	yujiepan	0	6.1k
mlx-community/gpt-oss-120b-mxfp4-bf16	quantized	mxfp4	117B	XXL	apache-2.0	mlx-community	3	5.9k
dealignai/GPT-OSS-120B-MLX-CRACK	quantized	MLX	117B	XXL	apache-2.0	dealignai	5	5.1k
exalandru/GPT-OSS-Coder-GGUF	finetune	GGUF	—	—	apache-2.0	exalandru	2	5k
michaelwaves/amoral-gpt-oss-120b-bfloat16	finetune	—	117B	XXL	apache-2.0	michaelwaves	1	4.4k

Repository changelog

17 commits

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

Repository changelog 17 commits

Date	Commit	SHA
2025-08-26	Update README.md	b5c939d
2025-08-13	Configure `<\|call\|>` as an EOS token (#105)	8b193b0
2025-08-13	Update README	5c98b88
2025-08-08	Update chat_template.jinja (#85)	bc75b44
2025-08-08	Update README.md	ef8a86a
2025-08-08	Update README.md	88fa8c2
2025-08-08	Update README.md	34a49b3
2025-08-07	Chat template cleanups + latest fixes (#76)	988e56b
2025-08-06	Remove chat_template.json (#58)	eabf0c5
2025-08-05	Update README	4c0fa49
2025-08-05	Update LICENSE	a9cc02b
2025-08-05	Update vLLM command	a040b35
2025-08-05	Add model card	acd9ca9
2025-08-05	Upload folder using huggingface_hub	f39accd
2025-08-05	Upload folder using huggingface_hub	ed282e6
2025-08-04	Upload folder using huggingface_hub	8c05803
2025-08-04	initial commit	62af5b1

All providers

27 providers · 27 endpoints

Where to run 27 providers · 27 endpoints

Hosted endpoints observed for gpt-oss-120b, 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: $0.03/M.

Provider	Model ref	Context	Quant	$/M in	$/M out	tok/s	TTFT	Tools	Status	Observed
DeepInfra hf, deepinfra, openrouter	openai/gpt-oss-120b	128K	bfloat16	$0.03	$0.14	76.4	228 ms	yes	live	2026-10-25
Near AI openrouter	openai/gpt-oss-120b	128K	fp4	$0.03	$0.15	—	—	yes	live	2026-10-24
Venice openrouter	openai/gpt-oss-120b	125K	int4	$0.03	$0.15	—	—	yes	live	2026-10-24
CoreWeave openrouter	openai/gpt-oss-120b	128K	fp4	$0.03	$0.17	—	—	yes	live	2026-10-24
DekaLLM openrouter	openai/gpt-oss-120b	128K	bf16	$0.03	$0.18	—	—	yes	live	2026-10-24
AkashML openrouter	openai/gpt-oss-120b	128K	bf16	$0.03	$0.19	—	—	yes	live	2026-10-24
Novita hf, openrouter	openai/gpt-oss-120b	128K	fp4	$0.05	$0.25	119.3	4648 ms	yes	live	2026-10-25
Crusoe openrouter	openai/gpt-oss-120b	128K	bf16	$0.05	$0.25	—	—	yes	live	2026-10-24
Mancer 2 openrouter	openai/gpt-oss-120b	128K	fp8	$0.05	$0.25	—	—	yes	live	2026-10-24
DigitalOcean openrouter	openai/gpt-oss-120b	125K	unknown	$0.06	$0.42	—	—	—	live	2026-10-24
OVHcloud hf	gpt-oss-120b	128K	—	$0.09	$0.47	37.5	927 ms	yes	live	2026-10-25
Google openrouter	openai/gpt-oss-120b	128K	unknown	$0.09	$0.36	—	—	—	status -5	2026-10-24
Nscale hf	openai/gpt-oss-120b	128K	—	$0.10	$0.40	122.8	348 ms	yes	live	2026-10-25
Baseten hf, openrouter	openai/gpt-oss-120b	125K	fp4	$0.10	$0.50	259.1	172 ms	yes	live	2026-10-25
Parasail parasail, openrouter	parasail-gpt-oss-120b	128K	fp4	$0.10	$0.75	—	—	yes	serverless	2026-10-24
SambaNova openrouter	openai/gpt-oss-120b	128K	unknown	$0.14	$0.95	—	—	yes	live	2026-10-24
Groq hf, openrouter	openai/gpt-oss-120b	128K	unknown	$0.15	$0.75	406.9	250 ms	yes	live	2026-10-25
Together AI hf, openrouter	openai/gpt-oss-120b	128K	unknown	$0.15	$0.60	159	536 ms	yes	live	2026-10-25
Fireworks AI hf	accounts/fireworks/models/gpt-oss-120b	128K	—	$0.15	$0.60	172.3	432 ms	yes	live	2026-10-25
Amazon Bedrock openrouter	openai/gpt-oss-120b	128K	unknown	$0.15	$0.60	—	—	—	live	2026-10-24
Nebius openrouter	openai/gpt-oss-120b	128K	fp4	$0.15	$0.60	—	—	yes	live	2026-10-24
SiliconFlow openrouter	openai/gpt-oss-120b	128K	fp8	$0.15	$0.60	—	—	—	live	2026-10-24
Phala openrouter	openai/gpt-oss-120b	128K	unknown	$0.15	$0.60	—	—	yes	live	2026-10-24
Mara openrouter	openai/gpt-oss-120b	128K	unknown	$0.15	$0.75	—	—	yes	live	2026-10-24
Scaleway hf	gpt-oss-120b	—	—	$0.17	$0.68	158.1	256 ms	yes	live	2026-10-25
Cerebras hf, openrouter	gpt-oss-120b	128K	fp16	$0.35	$0.75	178.6	4964 ms	yes	live	2026-10-25
Featherless hf	openai/gpt-oss-120b	—	—	—	—	—	—	—	live	2026-10-25

Articles and coverage

44 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.

Analysis5

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: ExpertCache – Run the Full GPT-OSS 120B (63GB) on a 16GB M1 Pro

github.com2026-08-08authorityHN 4 pts · 0 comments1,334 words

ExpertCache is an experimental page-aware Metal runtime that exposes only selected expert ranges to Metal through page-aligned direct host-memory views to run oversized sparse mixture-of-experts checkpoints. It targets the 63,387,346,208-byte GPT-OSS 120B MXFP4 GGUF checkpoint, achieving a 70% prefill speedup from 5.75 to 9.80 tokens/second on a 64 GiB M1 Max system. The approach avoids binding or copying the complete expert tensor into the active Metal working set.written, reviewed, cited

The 16 GiB M1 Pro path completed the seven-scenario 8K qualification in 8,249 seconds total.
The successful 16 GiB configuration explicitly disabled automatic fit and whole-file mmap prefetch.
ExpertCache-authored code is Apache-2.0, while the manuscript source and compiled paper are separately licensed CC BY 4.0.

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.

Running GPT-OSS-120B at 500 tokens per second on Nvidia GPUs

baseten.co2025-08-07authorityHN front pageHN 247 pts · 175 comments857 words

Baseten optimized GPT OSS 120B on NVIDIA GPUs to achieve SOTA latency and throughput at launch. The team reached speeds of over 650 tokens per second on Artificial Analysis and higher on OpenRouter. They selected Tensor Parallelism for better latency on Hopper and Blackwell architectures while using the TensorRT-LLM MoE Backend for Blackwell. Speculative decoding with Eagle is also being added to further improve performance.written, reviewed, cited

Optimized deployments parallelize the model across 4 or 8 GPUs for improved performance and throughput.
Tensor Parallelism offered better latency, while Expert Parallelism offered better system throughput.
The TensorRT-LLM MoE Backend adds improved CUDA kernels that outperform the previous Triton backend.

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

Community2

Show HN: I taught GPT-OSS-120B to see using Google Lens and OpenCV

news.ycombinator.com2026-02-11HN 43 pts · 31 comments1,569 words

GPT-OSS-120B runs on just 8GB VRAM & 64GB+ system RAM

old.reddit.com2025-08-11HN front pageHN 248 pts · 79 comments

All coverage tracked for this model (44) — 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
Is GPT-OSS Good? A Comprehensive Evaluation of OpenAI's ...	arxiv.org	—	2026-10-07	exa
gpt-oss-120b — Architecture, Evals & Inference Performance \| InferenceX	inferencex.semianalysis.com	—	2026-10-07	exa
Introducing OpenAI gpt-oss (20b & 120b) - Fireworks AI	fireworks.ai	—	2026-10-07	exa
gpt-oss-120b Coding Evaluation: New Top Open-Source Model	eval.16x.engineer	—	2026-10-07	exa
OpenAI's GPT-OSS-120B and GPT-OSS-20B Review - Kingy AI	kingy.ai	—	2026-10-07	exa
Show HN: ExpertCache – Run the Full GPT-OSS 120B (63GB) on a 16GB M1 Pro	github.com	2026-08-08	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
Show HN: We let GPT OSS 120B write and run Python and ARC AGI 2 jumped 4x	github.com	2026-02-11	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
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
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
What GPT-OSS leaks about OpenAI's training data	fi-le.net	2025-10-05	2026-10-07	hn
GPT-OSS Reinforcement Learning	docs.unsloth.ai	2025-09-27	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
Visualizing GPT-OSS-20B embeddings	melonmars.github.io	2025-08-17	2026-10-07	hn
GPT-OSS-120B (high): API Provider Performance Benchmarking	artificialanalysis.ai	2025-08-13	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
How Benchmaxxed is GPT-OSS-120B?	cmart.blog	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
OpenAI’s Open-Weight Models: A Technical Deep Dive into the gpt-oss-120b and gpt-oss-20b Models \| by Shubham \| Medium	shubh7.medium.com	2025-08-10	2026-10-07	exa
Curious about the training data of OpenAI's new GPT-OSS models? I was too	twitter.com	2025-08-09	2026-10-07	hn
Gptoss120B – OpenAI Open Source Large Language Model GPT-OSS-120B Official Site	gptoss120b.app	2025-08-09	2026-10-07	hn
Telnyx adds GPT-5 and Groq GPT-OSS-120B to Voice AI Agents	news.ycombinator.com	2025-08-08	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
AI medical diagnosis example with GPT-OSS 120B chain of thought	github.com	2025-08-07	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 120B Writes a Lisp in Go Fast	elite-ai-assisted-coding.dev	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 120B bad at producing linguistically correct German text	twitter.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
A first look at GPT-OSS-120B's coding ability	blog.brokk.ai	2025-08-05	2026-10-07	hn
GPT-OSS-120B Lisp interpreter in Go	gist.github.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

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

GLM-4.5-Air · 110B
same size band (XXL), also MoE, same task (text-generation), fewer providers, permissive license
Laguna-S-2.1 · 118B
same size band (XXL), also MoE, same task (text-generation), fewer providers
Ling-3.0-flash · 127B
same size band (XXL), 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-120b

Watch this model

/feeds/models/openai--gpt-oss-120b.xml(RSS 2.0) · every change to gpt-oss-120b
/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

GPT-OSS-120B runs on just 8GB VRAM & 64GB+ system RAM

Show HN: ExpertCache – Run the Full GPT-OSS 120B (63GB) on a 16GB M1 Pro

25 best of 44 tracked

Where to run it

Prices for the same weights vary 12×

Where it's used

Will it fit?

Fits two 80 GB GPUs at INT4; too big for one

How it scores

#6 of 13 on Aider Polyglot so far

Day-one receipts

The story so far

9 dated articles and 17 repo commits so far

What people built on it

504 derivatives on the Hub, led by 245 adapters

Reference

Benchmarks

Benchmarks 2 rows

Use-case evidence

What it is good at 5 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 17 commits

All providers

Where to run 27 providers · 27 endpoints

Articles and coverage

Official3

Analysis5

Community2

Papers

Papers 1

Similar models

Similar models

Sources

Sources

Follow gpt-oss-120b

Watch this model