gpt-oss-120b

Checkpoint identity

Official identifier

openai/gpt-oss-120bOpenAI gpt-oss-120b official model card (external source) · gpt-oss-120b checkpoint configuration (external source) · gpt-oss-120b Apache 2.0 license (external source)

Architecture

Mixture-of-experts reasoning language modelOpenAI gpt-oss-120b official model card (external source) · gpt-oss-120b checkpoint configuration (external source) · gpt-oss-120b Apache 2.0 license (external source)

Total parameters

117 billionPublished rounded counts; total stored weights determine the preliminary footprint, including all MoE experts.OpenAI gpt-oss-120b official model card (external source) · gpt-oss-120b checkpoint configuration (external source) · gpt-oss-120b Apache 2.0 license (external source)

Active parameters

5.1 billion per tokenActive parameters are a compute-related count, not the model's full stored weight count.OpenAI gpt-oss-120b official model card (external source) · gpt-oss-120b checkpoint configuration (external source) · gpt-oss-120b Apache 2.0 license (external source)

Modalities

text-input, text-outputOpenAI gpt-oss-120b official model card (external source) · gpt-oss-120b checkpoint configuration (external source) · gpt-oss-120b Apache 2.0 license (external source)

Context

131,072 tokensNative configuration where directly verified. Extensions, advertised K-labels and configuration differences are preserved in the notes below.OpenAI gpt-oss-120b official model card (external source) · gpt-oss-120b checkpoint configuration (external source) · gpt-oss-120b Apache 2.0 license (external source)

Release date

Exact release date not independently established in the reviewed artifactsOpenAI gpt-oss-120b official model card (external source) · gpt-oss-120b checkpoint configuration (external source) · gpt-oss-120b Apache 2.0 license (external source)

Download access

Public checkpoint referenceOpenAI gpt-oss-120b official model card (external source) · gpt-oss-120b checkpoint configuration (external source) · gpt-oss-120b Apache 2.0 license (external source)

License

Apache 2.0The weights include Apache 2.0 terms with redistribution notice, modification and patent-license conditions. The license does not itself establish suitability for a particular customer workload.OpenAI gpt-oss-120b official model card (external source) · gpt-oss-120b checkpoint configuration (external source) · gpt-oss-120b Apache 2.0 license (external source)

Theoretical weight-memory footprint

Using the publisher's approximately 117 billion total parameters at 16 bits per parameter gives 234 GB (217.9 GiB) of nominal weights.

This estimate excludes KV cache, activations, quantization metadata, non-weight buffers and serving overhead. It is not a tested GPU fit. MoE routing activates only part of the network per token while the complete expert weights still need storage.

Quantization

The publisher supplies MXFP4 MoE weights, while selected other modules remain unquantized. Compatible kernels and the correct format are required; plain 4-bit estimates are insufficient.

Serving considerations

• The name 120b is a product label; the card states approximately 117B total and 5.1B active parameters. Do not substitute 120B for the documented total.
• The config specifies 131,072 positions. The Harmony message format is required, including correct handling of reasoning and tool output.
• The publisher's single 80GB GPU statement is not Aguasto's measured capacity. Batch size, cache, context and backend affect fit; Aguasto has not tested or deployed this checkpoint.

Framework documentation

These are third-party documentation starting points. Support depends on the exact checkpoint, runtime version and hardware.