Across 12 comparable benchmarks, Opus 4.5 has better reported results on 11 and worse on 1.
Moderate evidence12 of 12 shared results countedHow we compare
Price basis: Standard rates. The price section below shows the applicable periods.
| Fact | Opus 4.5 | gpt-oss-20b |
|---|---|---|
| Context window | 200K tokens | 128K tokens |
| Input / 1M | $5.00 | $0.030 |
| Output / 1M | $25.00 | $0.140 |
| Blended (4:1) | $5.76 | $0.052 |
| Status | Available | Available |
| Released | November 25, 2025 | August 6, 2025 |
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
Opus 4.5 ahead on 2 of 2 counted
gpt-oss-20b ahead on 1 of 1 counted
Opus 4.5 ahead on 2 of 2 counted
Opus 4.5 ahead on 1 of 1 counted
Opus 4.5 ahead on 3 of 3 counted
No shared comparable results for Abstract reasoning, Research & search, Long context, Multimodal, Cybersecurity.
Published scores, with matching modes where available. Gaps show Opus 4.5 relative to gpt-oss-20b. Reading the results
| Benchmark | Opus 4.5 | gpt-oss-20b | Gap |
|---|---|---|---|
| SWE-bench VerifiedRepository Engineering · Accuracy | 80.9Extended · tools | 34.0≠Thinking · no tools | +46.9 pts#9 vs #103 of 108 |
| Terminal Bench HardAgentic Development · Accuracy | 47.0Thinking · tools | 10.6≠Thinking High · tools | +36.4 pts#15 vs #115 of 157 |
| AIME2025Mathematics · Accuracy | 91.3Thinking · no tools | 98.7≠Thinking · tools | −7.4 pts#34 vs #13 of 86 |
| CritPtScientific Reasoning · Score | 4.6Thinking · no tools | 1.4≠Thinking High · no tools | 3.29×#73 vs #96 of 134 |
| GPQA DiamondScientific Reasoning · Accuracy | 87.0Extended · no tools | 71.5≠Thinking · no tools | +15.5 pts#70 vs #130 of 200 |
| Pinch BenchTool Orchestration · Accuracy | 87.2Extended · tools | 66.0≠Thinking · tools | +21.2 pts#9 vs #35 of 38 |
| Creative WritingWriting | 1686Normal · no tools | 665Normal · no tools | 2.54×#35 vs #119 of 121 |
| τ²-BenchService Workflows · Accuracy | 82.0Extended · tools | 47.7≠Thinking · tools | +34.3 pts#13 vs #34 of 40 |
| τ³-BankingService Workflows · Score | 24.7Thinking High · tools | 7.0Thinking High · tools | 3.53×#48 vs #99 of 120 |
| HLEKnowledge Exams · Accuracy | 43.2Extended · tools | 17.3≠Thinking · tools | +25.9 pts#72 vs #114 of 151 |
| MMLU-ProKnowledge Exams · Accuracy | 90.0Extended · no tools | 74.0≠Thinking · no tools | +16.0 pts#1 vs #84 of 153 |
| ECICapability Indices · ECI score (capability index, higher is better) | 150Reported best | 138Reported best | 1.09×#47 vs #106 of 168 |
Repository Engineering · Accuracy
Agentic Development · Accuracy
Mathematics · Accuracy
Scientific Reasoning · Score
Scientific Reasoning · Accuracy
Tool Orchestration · Accuracy
Writing
Service Workflows · Accuracy
Service Workflows · Score
A dash means no result is available.
| Benchmark | Opus 4.5 | gpt-oss-20b |
|---|---|---|
| AIME 2024 | — | 96.0 |
| AIME 2026 | 93.3 | — |
| FrontierMath | 20.7 | — |
| FrontierMath - Tier 4 | 4.2 | — |
| FrontierMath Tier 4 v2 | 4.9 | — |
| FrontierMath v2 | 34.4 | — |
| HLE-Verified | 38.8 | — |
| IMO-ProofBench Advanced | 23.8 | — |
| MMLU | — | 85.3 |
| Riemann-bench | 11.2 | — |
| SciCode | — | 38.9 |
| APEX-SWE | 38.7 | — |
| CodeForces | — | 2516 |
| GSO | 26.5 | — |
| LiveCodeBench | 87.1 | — |
| SWE-bench | 52.6 | — |
| SWE-bench Multilingual | 70.7 | — |
| SWE-Bench Pro - Commercial | 23.4 | — |
| SWE-Bench Pro - Public | 45.9 | — |
| Terminal Bench 2.0 | 59.3 | — |
| Vibe Code Bench v1.1 | 20.6 | — |
| WeirdML v2 | 63.7 | — |
| WeirdML v3 | 0.065 | — |
| Claw Bench | 91.5 | — |
| MCP-Atlas | 69.8 | — |
| PinchBench v2 | — | 36.3 |
| LiveBench | 72.6 | — |
| METR Time Horizons v1.1 | 293 | — |
| Text Arena (Coding) | 1512 | — |
| IF Bench | 58.0 | — |
| EBR-bench | 14.3 | — |
| LongBench v2 | 64.4 | — |
| GDP.pdf | — | 2.0 |
| GeoBench ACW | 75.0 | — |
| MMMU | 80.7 | — |
| MMMU-Pro | 74.0 | — |
| VPCT | 40.0 | — |
| SAGE | 52.1 | — |
| τ²-Bench - Telecom | 92.3 | — |
| ARC-AGI-1 | 80.0 | — |
| ARC-AGI-2 | 37.6 | — |
| BALROG | 43.5 | — |
| SimpleBench | 62.0 | — |
| Vending-Bench 2 | 4967 | — |
| BrowseComp | — | 28.3 |
| MedCode | 49.2 | — |
| MedScribe | 85.3 | — |
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%
Opus 4.5 costs10,977%morethan gpt-oss-20b per million tokens — 110.8× the price.Using standard rates at this token mix.
Opus 4.5: Standard pricing. Anthropic official pricing (platform.claude.com/docs/en/about-claude/pricing) verified 2026-11-13
gpt-oss-20b: Standard pricing. DeepInfra official models API (openai/gpt-oss-20b) verified 2026-11-13
| Price / 1M tokens | Opus 4.5 | gpt-oss-20b | Ratio |
|---|---|---|---|
| Input | $5.00 | $0.03 | 167× |
| Output | $25.00 | $0.14 | 179× |
| Cached input | $0.50 | — | — |
| Cache write | — | — | — |
| Blended | $5.76 | $0.052 | 111× |
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 | Opus 4.5 | gpt-oss-20b |
|---|---|---|
| ARC-AGI-1 | Not reported | |
| ARC-AGI-2 | Not reported | |
| WeirdML v3 | $8.7034 / task Average · Official score (0-1; higher is better): 0.065 Thinking High (With Tools) | Not reported |
| Attribute | Opus 4.5 | gpt-oss-20b |
|---|---|---|
| Availability | Available | Available |
| Provider | Anthropic | OpenAI |
| Model type | Reasoning model | Reasoning model |
| First release | November 25, 2025 | August 6, 2025 |
| Context window | 200K tokens | 128K tokens |
| Max output | 64K tokens | 128K tokens |
| Total parameters | — | 21B |
| Active parameters | — | 3.6B |
| Architecture | Undisclosed | Mixture of Experts (MoE) |
| Documented capabilities | Reasoning, Multilingual | Reasoning, Multilingual |
| Knowledge cutoff | 2025-03 | — |
| Input modalities | Text, Image | Text |
| Output modalities | Text | Text |
| Weights | Closed / API only | Open weights |
| Licence | — | Apache 2.0 |
| Commercial use | Proprietary | 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 Opus 4.5 or gpt-oss-20b? 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.