gpt-oss-120b
An open-weight large language model with 120 billion parameters released by OpenAI. It offers broad general-purpose capabilities and demonstrates strong performance across comprehensive benchmark evaluations.
Core Facts
Timeline (last 90 days)
The researchers validated the CoM method on two open-weight models, gemma-4-31b-it and gpt-oss-120b.
The same open model, gpt-oss-120b, scored anywhere from 63.9% to 78.0% on the same 800 tasks depending on the four different hosting providers.
Two open-weight models showed nuclear escalation: gpt-oss-120b triggered midnight once across the three tests, and DeepSeek V4.1 Flash triggered midnight twice across the three tests.
gpt-oss-120b fabricated a complete biography in 5 out of 5 attempts, including invented names, employers and résumés.
The reasoning layer of MedScribe AI uses openai/gpt-oss-120b running through Groq, achieving fast and free inference, and patient X-ray data is not sent to any proprietary server.
gpt-oss-120b has about 117B total parameters, but each token activates only about 5.1B parameters.
The same parameter name reasoning_effort behaves differently across models from different vendors: in gpt-5.4-mini the median output tokens rose from 18 at none to 254 at high (×14.1); gpt-oss-120b rose from refusal to 1,340 (×4.6); claude-haiku-5.5 rose from 150 to 298 (×2.0); gemini-3.7-flash rose from 453 to 1,013 (×2.2).
gpt-oss-120b spent about 25 times the cost per correct answer on the distractor counting task, while its accuracy was 7.5 percentage points lower (not significant).
All incorrect distractor counting answers were without exception overestimates, and the token counts of incorrect answers were clearly higher than those of correct answers: for gpt-5.4-mini at high, the median token count of incorrect answers was 412 vs 91.5 for correct answers; for gpt-oss-120b, incorrect answers were 5,776 tokens vs 747 tokens for correct answers.
gpt-oss-120b directly returns an HTTP 400 error when reasoning_effort='none', refusing that value.
20 more timeline events
All Facts (20)
The models tested include Google's Gemini 3.7 Flash, Z.ai's GLM-5, Alibaba's Qwen 3 235B A22B Instruct, and OpenAI's gpt-oss-120b, all running on Kaggle.
65%UnverifiedThe researchers validated the CoM method on two open-weight models, gemma-4-31b-it and gpt-oss-120b.
50%UnverifiedThe same open model, gpt-oss-120b, scored anywhere from 63.9% to 78.0% on the same 800 tasks depending on the four different hosting providers.
50%UnverifiedTwo open-weight models showed nuclear escalation: gpt-oss-120b triggered midnight once across the three tests, and DeepSeek V4.1 Flash triggered midnight twice across the three tests.
50%UnverifiedThe reasoning layer of MedScribe AI uses openai/gpt-oss-120b running through Groq, achieving fast and free inference, and patient X-ray data is not sent to any proprietary server.
50%Unverifiedgpt-oss-120b has about 117B total parameters, but each token activates only about 5.1B parameters.
50%Unverifiedgpt-oss-120b spent about 25 times the cost per correct answer on the distractor counting task, while its accuracy was 7.5 percentage points lower (not significant).
50%UnverifiedAll incorrect distractor counting answers were without exception overestimates, and the token counts of incorrect answers were clearly higher than those of correct answers: for gpt-5.4-mini at high, the median token count of incorrect answers was 412 vs 91.5 for correct answers; for gpt-oss-120b, incorrect answers were 5,776 tokens vs 747 tokens for correct answers.
50%Unverifiedgpt-oss-120b directly returns an HTTP 400 error when reasoning_effort='none', refusing that value.
50%UnverifiedCode replay inherits errors from the extraction stage; in the cluster sufficiency test, gpt-oss-120b's pass rate actually dropped from 56% to 40%.
50%UnverifiedFor gpt-oss-120b on the same 48 core families, the pass rate was approximately 63% on the Kaggle platform and as high as 98% on the cluster; the discrepancy stems from hosted answers being truncated mid-word, and on the 31 families where hosted answers were complete, both platforms passed 30.
50%Unverifiedgpt-oss-120b scored 0.625, lower than gpt-oss-20b's 0.720, indicating that a larger model does not necessarily mean a safer one.
50%UnverifiedLarger, newer models do not necessarily perform better: gpt-oss-120b (0.740) scored lower than gpt-oss-20b (0.761); DeepSeek-R1, costing $1.94, scored lower than Gemma 4 31B, which cost only 13 cents.
50%UnverifiedFor false information such as incorrectly typed dates, the correct action is to 'erase' rather than 'change'; gpt-oss-120b performed the correction correctly only 42% of the time, and Qwen3 235B only half the time.
50%Unverifiedgpt-oss-120b retained 11 of 12 bookings but refused zero injections, writing the fake address from a phishing email into the passport; Qwen3 235B also stored the same address in the voucher field.
50%Unverifiedgpt-oss-120b scored 0.991 on memory read but only 0.740 on write, ranking last in write; DeepSeek-R1 scored 0.997 on read and 0.863 on write.
50%UnverifiedWhen gpt-oss is loaded with load_in_4bit = False, MXFP4 experts are retained, allowing gpt-oss-120b to be trained on a single B200
50%UnverifiedOn broader comprehensive benchmark aggregate scores, the 120B model still leads TwIL-LM3.
50%UnverifiedIn terms of throughput, TwIL-LM3 reaches 32.9 answers/sec, while gpt-oss-120b achieves 12.6 answers/sec.
50%UnverifiedOn the Rule Induction task, TwIL-LM3 scored 96.4 while gpt-oss-120b scored 65.2.
50%