Across 13 comparable benchmarks, Nemotron 3 Ultra has better reported results on 13 and worse on 0.
Moderate evidence13 of 13 shared results countedHow we compare
Price basis: Standard rates. The price section below shows the applicable periods.
| Fact | gpt-oss-20b | Nemotron 3 Ultra |
|---|---|---|
| Context window | 128K tokens | 1M tokens |
| Input / 1M | $0.030 | $0.600 |
| Output / 1M | $0.140 | $2.40 |
| Blended (4:1) | $0.052 | $0.614 |
| Status | Available | Available |
| Released | August 6, 2025 | June 4, 2026 |
Blended is 4:1 input to output with 90% cache hits at listed rates — set your own mix.
Mean of 6 shared benchmarks grouped by capability on a 0–100 scale. How averages are built
Nemotron 3 Ultra ahead on 2 of 2 counted
Nemotron 3 Ultra ahead on 2 of 2 counted
No fixed-scale average — see the rows in Benchmark results
Nemotron 3 Ultra ahead on 2 of 2 counted
Nemotron 3 Ultra ahead on 1 of 1 counted
Nemotron 3 Ultra ahead on 1 of 1 counted
Nemotron 3 Ultra ahead on 2 of 2 counted
No fixed-scale average — see the rows in Benchmark results
No shared comparable results for Abstract reasoning, Mathematics, Long context, Cybersecurity.
Published scores, with matching modes where available. Gaps show gpt-oss-20b relative to Nemotron 3 Ultra. Reading the results
| Benchmark | gpt-oss-20b | Nemotron 3 Ultra | Gap |
|---|---|---|---|
| SWE-bench VerifiedRepository Engineering · Accuracy | 34.0Thinking · no tools | 70.7≠Thinking · tools | −36.7 pts#103 vs #58 of 108 |
| Terminal Bench HardAgentic Development · Accuracy | 10.6Thinking High · tools | 36.4≠Thinking · tools | −25.8 pts#115 vs #45 of 157 |
| CritPtScientific Reasoning · Score | 1.4Thinking High · no tools | 3.1≠Thinking · no tools | 0.45×#96 vs #78 of 134 |
| SciCodeScientific Computing · Score | 38.9Thinking High · no tools | 44.6≠Thinking · no tools | 0.87×#79 vs #66 of 98 |
| Pinch BenchTool Orchestration · Accuracy | 66.0Thinking · tools | 90.0Thinking · tools | −24.0 pts#35 vs #2 of 38 |
| PinchBench v2Tool Orchestration · Average score (%) | 36.3Reported best | 89.9Reported best | −53.6 pts#48 vs #4 of 51 |
| BrowseCompFact Finding · Accuracy | 28.3Thinking · tools | 44.4≠Thinking · tools + web | −16.1 pts#58 vs #51 of 61 |
| GDP.pdfDocuments & Charts · Task pass rate (%) | 2.0Thinking High · no tools | 5.0≠Thinking · no tools | −3.0 pts#84 vs #75 of 96 |
| Creative WritingWriting | 665Normal · no tools | 1692Normal · no tools | 0.39×#119 vs #33 of 121 |
| τ³-BankingService Workflows · Score | 7.0Thinking High · tools | 22.6≠Thinking · tools | 0.31×#99 vs #54 of 120 |
| HLEKnowledge Exams · Accuracy | 17.3Thinking · tools | 37.4Thinking · tools | −20.1 pts#114 vs #80 of 151 |
| MMLU-ProKnowledge Exams · Accuracy | 74.0Thinking · no tools | 86.8Thinking · no tools | −12.8 pts#84 vs #19 of 153 |
| ECICapability Indices · ECI score (capability index, higher is better) | 138Reported best | 146Reported best | 0.94×#106 vs #70 of 168 |
Repository Engineering · Accuracy
Agentic Development · Accuracy
Tool Orchestration · Accuracy
Tool Orchestration · Average score (%)
Fact Finding · Accuracy
Documents & Charts · Task pass rate (%)
Writing
Service Workflows · Score
A dash means no result is available.
| Benchmark | gpt-oss-20b | Nemotron 3 Ultra |
|---|---|---|
| AIME 2024 | 96.0 | — |
| AIME2025 | 98.7 | — |
| GPQA | — | 87.0 |
| GPQA Diamond | 71.5 | — |
| IMO-AnswerBench | — | 92.3 |
| MMLU | 85.3 | — |
| AA Cyber Index | — | 13.1 |
| Code Migration | — | 4.9 |
| CodeForces | 2516 | — |
| FrontierCode 1.1 | — | 13.6 |
| LiveCodeBench | — | 89.0 |
| SWE-bench Multilingual | — | 67.7 |
| Terminal-Bench 2.1 | — | 56.4 |
| Terminal-Bench 4.0 | — | 0.5 |
| Vibe Code Bench v1.1 | — | 7.6 |
| APEX-Agents | — | 22.7 |
| MCP-Atlas | — | 63.1 |
| Toolathlon-Verified | — | 34.3 |
| CWE-bench v0 | — | 18.5 |
| AA Intelligence Index (historical versions) | — | 38.3 |
| AA Intelligence Index v5.0.1 | — | 11.8 |
| LiveBench | — | 67.4 |
| LiveBench 2.0 | — | 40.8 |
| IF Bench | — | 81.7 |
| LongBench v2 | — | 61.9 |
| AA-AnalystAgent | — | 6.3 |
| EMB (Excel Modeling) | — | 31.7 |
| Harvey Lab-AA | — | 81.7 |
| Legal Research Bench | — | 15.4 |
| τ²-Bench | 47.7 | — |
| SimpleBench | — | 41.7 |
| MedCode | — | 38.6 |
Time-of-day providers use peak-hour rates; fixed-price providers use standard rates. Off-peak discounts remain in the published details.
Drag the handle above the bar for cache hits, the one below it for the input : output mix · 5% steps
Blended price = effective input rate at 90% cache hits × 80% + output rate × 20%
gpt-oss-20b costs92%lessthan Nemotron 3 Ultra per million tokens — 11.8× cheaper.Using standard rates at this token mix.
gpt-oss-20b: Standard pricing. DeepInfra official models API (openai/gpt-oss-20b) verified 2026-11-13
Nemotron 3 Ultra: Standard pricing. Fireworks official Serverless pricing (fireworks/nemotron-3-ultra-nvfp4) verified 2026-11-13
| Price / 1M tokens | gpt-oss-20b | Nemotron 3 Ultra | Ratio |
|---|---|---|---|
| Input | $0.03 | $0.60 | 0.05× |
| Output | $0.14 | $2.40 | 0.06× |
| Cached input | — | $0.12 | — |
| Cache write | — | — | — |
| Blended | $0.052 | $0.614 | 0.08× |
Estimates use the selected token mix. Cache writes, tool fees and long-context surcharges are excluded.
Reported cost per task, paired with the score from the same run. Settings and sources may differ; costs do not affect the verdict.
| Benchmark | gpt-oss-20b | Nemotron 3 Ultra |
|---|---|---|
| AA Intelligence Index v5.0.1 | Not reported | |
| CWE-bench v0 | Not reported | |
| LiveBench 2.0 | Not reported | $0.5057 / task Average · Overall score (equal-weight mean of 7 suite scores, 0-100): 40.8 Max (task-specific tools) |
| Attribute | gpt-oss-20b | Nemotron 3 Ultra |
|---|---|---|
| Availability | Available | Available |
| Provider | OpenAI | NVIDIA |
| Model type | Reasoning model | Reasoning model |
| First release | August 6, 2025 | June 4, 2026 |
| Context window | 128K tokens | 1M tokens |
| Max output | 128K tokens | — |
| Total parameters | 21B | 550B |
| Active parameters | 3.6B | 55B |
| Architecture | Mixture of Experts (MoE) | Mixture of Experts (MoE) |
| Documented capabilities | Reasoning, Multilingual | Reasoning, Multilingual, Tool use, Long context |
| Knowledge cutoff | — | 2026-05 |
| Input modalities | Text | Text |
| Output modalities | Text | Text |
| Weights | Open weights | Open weights |
| Licence | Apache 2.0 | — |
| Commercial use | Commercial use allowed | Commercial use allowed |
These are reported-score comparisons, not controlled head-to-head tests.
What people found comparing these models, or using either one. Comments also appear on the models' own pages.
Used gpt-oss-20b or Nemotron 3 Ultra? Sign in with Google to share what worked, what did not, and which API or setup you tested.
No community comments on either model yet.
Alternatives with the most benchmark overlap for this matchup.