model
gpt-oss 120B llm
A model record: onboarded once for the project and used by any eval's arms.
- weights
- open
- licence
- Apache-2.0
- aliases
gpt-oss:120beval-gpt-oss-120b- settings
reasoning_effortlow reasoning cannot be turned off: with none it still reasoned on every machine, and the hosted route answers "Reasoning is mandatory" (onboarding, 2026-10-31); low is the least it takes- onboarding
- vetted 2026-10-31
- best result per goal
- #16 of 55 Triage decisions on Orrery's error signals
01deployments and their onboarding
x3 · x3 · ollama 0.40.13 · gpt-oss:120b · onboarding pass 2026-10-31 · 26.8 tokens/s · 1 accepted
- where
- x3
- server or device
- ollama 0.40.13
- build
gpt-oss:120b714d86958b6e870a9320e78126ac1ce3477727a1cd9bf7194b8ace0d84c62fd8MXFP4- matches the record
- yes
- checked
- 2026-10-31
- onboarded on
- x3
| probe | measurement | flag |
|---|---|---|
| endpoint | provider openai · base url http://127.0.0.1:11434/v1 · model gpt-oss:120b | — |
| digest found | 714d86958b6e870a9320e78126ac1ce3477727a1cd9bf7194b8ace0d84c62fd8 | — |
| default | s 18.43 · tokens out 98 · reasoned: yes · finish stop | — |
| chosen | s 1.83 · tokens out 49 · reasoned: yes · finish stop · settings (reasoning effort low) | — |
| tokens per s | 26.8 | — |
| deterministic | yes | — |
| json valid | yes | — |
Accepted: reasoning: it still reasons with the record's settings (it reasons whatever it is asked: with none and with low on every machine and the hosted route, which answers "Reasoning is mandatory" (onboarding, 2026-10-31); low is the least)
dgx · dgx · ollama 0.40.13 · gpt-oss:120b · onboarding pass 2026-10-31 · 34.0 tokens/s · 1 accepted
- where
- dgx
- server or device
- ollama 0.40.13
- build
gpt-oss:120b714d86958b6e870a9320e78126ac1ce3477727a1cd9bf7194b8ace0d84c62fd8- matches the record
- yes
- checked
- 2026-10-31
- onboarded on
- dgx
| probe | measurement | flag |
|---|---|---|
| endpoint | provider openai · base url http://100.69.49.126:11434/v1 · model gpt-oss:120b | — |
| digest found | 714d86958b6e870a9320e78126ac1ce3477727a1cd9bf7194b8ace0d84c62fd8 | — |
| default | s 14.15 · tokens out 98 · reasoned: yes · finish stop | — |
| chosen | s 1.47 · tokens out 50 · reasoned: yes · finish stop · settings (reasoning effort low) | — |
| tokens per s | 34.0 | — |
| deterministic | yes | — |
| json valid | yes | — |
Accepted: reasoning: it still reasons with the record's settings (it reasons whatever it is asked: with none and with low on every machine and the hosted route, which answers "Reasoning is mandatory" (onboarding, 2026-10-31); low is the least)
hosted · homelab LiteLLM (the fleet's gateway) · eval-gpt-oss-120b · onboarding pass 2026-10-31 · 28.1 tokens/s · 1 accepted
- where
- homelab LiteLLM (the fleet's gateway)
- build
eval-gpt-oss-120b- matches the record
- yes
- checked
- 2026-10-31
- onboarded on
- homelab LiteLLM (the fleet's gateway)
| probe | measurement | flag |
|---|---|---|
| endpoint | provider litellm · base url http://homelab:4001/v1 · model eval-gpt-oss-120b | — |
| digest found | None | — |
| default | s 2.28 · tokens out 83 · reasoned: yes · finish stop | — |
| chosen | s 1.35 · tokens out 38 · reasoned: yes · finish stop · settings (reasoning effort low) | — |
| tokens per s | 28.1 | — |
| deterministic | no | — |
| json valid | yes | — |
Accepted: reasoning: it still reasons with the record's settings (it reasons whatever it is asked: with none and with low on every machine and the hosted route, which answers "Reasoning is mandatory" (onboarding, 2026-10-31); low is the least)
What it can do, across its deployments: reasons: yes · json mode: yes · reasoning off None
02settings history
| when | what changed | why |
|---|---|---|
| 2026-10-31 | reasoning_effort: (server default) → none | it reasoned by default on the onboarding probe (2026-10-31); off unless a task needs it (RUNBOOK §16): left at the server's default, one model reasons at length and the next not at all |
| 2026-10-31 | reasoning_effort: none → low | reasoning cannot be turned off: with none it still reasoned on every machine, and the hosted route answers "Reasoning is mandatory" (onboarding, 2026-10-31); low is the least it takes |
03arms
| arm | deployment | goal | its settings | arm smoke | result |
|---|---|---|---|---|---|
| triage_hosted_gpt_oss_120b_v1 · rescored 5f2c057c orrery_triage_v1_labelled | hosted | Triage decisions on Orrery's error signals | — | none | triage_cost 0.677 |
| triage_hosted_gpt_oss_120b_v2 · rescored 5f2c057c orrery_triage_v1_labelled | hosted | Triage decisions on Orrery's error signals | max_tokens 1024 | pass 2026-10-31 | triage_cost 0.631 |
| triage_local_dgx_gpt_oss_120b_v1 · rescored 5f2c057c orrery_triage_v1_labelled | dgx | Triage decisions on Orrery's error signals | call_timeout_s 600 | none | triage_cost 0.692 |
| triage_local_dgx_gpt_oss_120b_v2 · rescored 5f2c057c orrery_triage_v1_labelled | dgx | Triage decisions on Orrery's error signals | max_tokens 1024, call_timeout_s 600 | check 2026-10-31 | triage_cost 0.662 |