GPT-OSS 20B gpt-oss-20b
openai · gpt-oss reasoning efficient
#75 overall
- params
- 21B (A3.6B)
- arch
- moe
- context
- 128k
- license
- apache-2.0 open weights
- released
- Aug 2025
- reasoning
- yes
- train compute
- 5.5e+23 FLOP
- downloads/30d
- 5.8M
# architecture
- layers
- 24
- d_model
- 2880
- heads
- 64q / 8kv
- head dim
- 64
- experts
- top-4 of 32
- vocab
- 201088
- family
- gpt_oss
signal path of one layer, generated from the registry's structured fields — dimension lines quote the model's real numbers; an MoE trace forks at the router, a dense trace runs straight through. Geometry fields come from the repo's config.json.
# why ranked
Overall: #75 — score 30.9 ● low — a single signal; treat with caution
| signal | weight | input (0–100) |
|---|---|---|
| aa_intelligence | 0.70 | 23.6 |
| bench_composite | 0.30 | 48.0 |
benchmark panel evidence:
| complete panel | score (0–100) |
|---|---|
| aa-coding-v1 | 32.2 |
| aa-math-v1 | 63.8 |
Coding: provisional — score 22.1 ● low — a single signal; treat with caution
provisional: too few independent signals for a numbered position — the score is shown, but this model sorts after every ranked model.
| signal | weight | input (0–100) |
|---|---|---|
| aa_coding | 0.30 | 22.1 |
| aider_polyglot | 0.30 | — |
| swe_bench_verified | 0.40 | — |
missing signals are dropped and the remaining weights renormalized — never imputed.
# trend
methodology changed during this history window; score movement across that boundary is not model movement. See methodology v6.
# benchmarks
| benchmark | score | source | date |
|---|---|---|---|
| AA Coding Index | 20.7 | Artificial Analysis | — |
| AA Intelligence Index | 9.0 | Artificial Analysis | — |
| AA Math Index | 89.3 | Artificial Analysis | — |
| Codeforces | 0.7 / 3000 | LLM Stats | — |
# where to run
| provider | quant | ctx | $/M in | $/M out | $/M cache | price src | tps | uptime | ✓ |
|---|---|---|---|---|---|---|---|---|---|
| Featherless | — | — | — | — | via hfrouter | — | — | ||
| Darkbloom | fp8 | 128k | $0.02 | $0.09 | $0.01 | via openrouter | — | 100.0% | |
| AkashML | fp4 | 128k | $0.02 | $0.10 | — | via openrouter | — | 100.0% | |
| CoreWeave | fp4 | 128k | $0.03 | $0.13 | $0.03 | via openrouter | — | 100.0% | |
| Wandb | 131k | $0.03 | $0.13 | — | via litellm | — | — | ||
| DekaLLM | bf16 | 128k | $0.03 | $0.14 | $0.03 | via openrouter | — | 100.0% | |
| DeepInfra | bf16 | 128k | $0.03 | $0.14 | — | deepinfra | — | 99.5% | ✓ |
| Parasail | fp4 | 128k | $0.03 | $0.15 | $0.02 | via openrouter | — | 99.9% | |
| Amazon Bedrock | unknown | 128k | $0.03 | $0.15 | — | bedrock | — | 100.0% | ✓ |
| Amazon Bedrock | unknown | 128k | $0.03 | $0.15 | — | bedrock | — | 100.0% | ✓ |
| Novita | 128k | $0.04 | $0.15 | — | via litellm | — | — | ||
| SiliconFlow | fp8 | 128k | $0.04 | $0.18 | — | openrouter | — | 88.7% | ✓ |
| Inferenceapis | 128k | $0.04 | $0.19 | — | via litellm | — | — | ||
| OVHcloud | 128k | $0.05 | $0.18 | — | ovhcloud | 54 | — | ✓ | |
| Together | 128k | $0.05 | $0.20 | — | via litellm | — | — | ||
| Nscale | 128k | $0.05 | $0.20 | — | via hfrouter | 164 | — | ||
| unknown | 128k | $0.07 | $0.25 | — | via openrouter | — | — | ||
| Tensormesh | 128k | $0.07 | $0.28 | — | via litellm | — | — | ||
| Bedrock Mantle | 128k | $0.07 | $0.30 | — | via litellm | — | — | ||
| Fireworks | 128k | $0.07 | $0.30 | $0.04 | via requesty | — | — | ||
| Groq | unknown | 128k | $0.07 | $0.30 | — | groq | — | 99.7% | ✓ |
| Replicate | — | $0.09 | $0.36 | — | via litellm | — | — | ||
| Cloudflare | 125k | $0.20 | $0.30 | — | via modelsdev | — | — |
sorted by blended price ((3·input + output) / 4 per 1M) · ✓ = the provider's own catalog confirms the offer · "via …" prices are what the aggregator routing the offer charges, not the provider's own list price
source aliases
- aa
gpt-oss-20b- arena
gpt-oss-20b- openrouter
openai/gpt-oss-20b