Up to date
openai Apache 2.0 out 15 months ago
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
Why did the open-weights model cross the road?
To get to the other side of the dataset.
__/\\__
/ \
/ O O \
/ ^ \
/__________\
|||
|||
/ \
~~~ ~~~~~ ~~~ ~~~~ 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 →
Picked by rule from 44 tracked pieces: the lab's own post, the most shared, a field report, the latest.
OpenAI released gpt-oss-120b and gpt-oss-20b, their first open-weight models since GPT-2 in 2019.
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.
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.
Live hosted endpoints, cheapest first, USD per million tokens. Ties break on output price, then provider name.
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.
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).
Top Spaces running it: hadadxyz/ai ♥ 819 · ginipick/FLUXllama ♥ 561 · smolagents/ml-intern ♥ 423 · burtenshaw/karpathy-llm-council ♥ 292
Weights only, an estimate: leave room for the KV cache. Method
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
Third-party measurements only; author-reported numbers stay in Reference, marked as such.
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
The probe replies as received on 2026-10-07 from parasail, graded by code. Receipts, not a score.
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 amber-falcon-7731
{"city": "Lisbon", "unit": "c"} {
"name": "Foo-9B-Instruct",
"license": "Apache-2.0",
"parameters_b": 9.2,
"context_tokens": 131072,
"languages": ["English", "French", "German", "Japanese"]
} {
"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."
} {
"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"
} Weights, every dated article, and repo commits on one rail.
Teal is the weights (2025-08-04), amber triangles are 9 dated articles (each links to it), grey ticks are repo commits.
504 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.
From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.
Everything else, folded: the full tables, every source.
| 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 |
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.
| 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 |
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.
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.
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.
| 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
quantized: 134finetune: 124adapter: 245merge: 1
| 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 |
| 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 |
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 |
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.
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).
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
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.