vs
13:3
overall capability wins (1 ties across 17 scored). gpt-oss-120B leads overall.
- Reasoning6:1
- Coding3:2
- Arena1:0
- Tool use1:0
- Composite indices2:0
- Pricing0:3
- Speed0:2
- Specs2:0
Verdict
gpt-oss-120B leads coding (widest gap: +19.2 on SWE-Rebench); gpt-oss-120B ranks higher on LMArena (+35 Elo); gpt-oss-120B allows ~2.0x more max output (33K vs 16K).
- gpt-oss-120B leads coding (widest gap: +19.2 on SWE-Rebench)
- gpt-oss-120B ranks higher on LMArena (+35 Elo)
- gpt-oss-120B allows ~2.0x more max output (33K vs 16K)
- gpt-oss-20B is ~3.0x cheaper on a blended token basis than gpt-oss-120B
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- gpt-oss-120B
- $0.30
- gpt-oss-20B
- $0.10
gpt-oss-20B is about 3.0x cheaper on this workload.
Price per 1M tokens
Full comparison
Organization
OpenAI
OpenAI
Family
gpt-oss
gpt-oss
License
Apache 2.0
Apache 2.0
Open weights
Yes
Yes
Release
Aug 5, 2025
Aug 5, 2025
Knowledge cutoff
-
-
API / provider
self-host / partners
self-host / partners
Modalities
text → text
text → text
Specs
Context window
131K
131K
Max output
33K
16K
Parameters
117B
21B
Pricing
Input $/1M
$0.15
$0.05
Output $/1M
$0.60
$0.20
Blended $/1M (3∶1)
$0.26
$0.09
Speed
tok/s
70
120
TTFT (s)
0.4
0.2
Reasoning
MMLU-Pro
—
—
GPQA Diamond
78.2
68.8
Humanity's Last Exam
19.6
11.0
AIME 2025
92.5
91.7
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
21.8
16.0
AA-LCR v1.1
52.0
34.7
CritPt
1.1
1.4
MMMU-Pro
—
—
IFBench
69.0
65.1
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
26.0
60.7
SWE-bench Pro
16.2
—
SWE-bench Multilingual
—
—
LiveCodeBench
—
—
Terminal-Bench 2.1
26.2
13.9
Aider Polyglot
41.8
34.2
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
34.0
38.9
CursorBench
—
—
SWE-Rebench
27.3
8.1
NL2Repo-Bench
—
—
DeepSWE
—
—
WebDev Arena
—
—
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
1,352
1,317
| Metric | gpt-oss-120B | gpt-oss-20B | Delta |
|---|---|---|---|
| Identity | |||
| Organization | OpenAI | OpenAI | — |
| Family | gpt-oss | gpt-oss | — |
| License | Apache 2.0 | Apache 2.0 | — |
| Open weights | Yes | Yes | — |
| Release | Aug 5, 2025 | Aug 5, 2025 | — |
| Knowledge cutoff | - | - | — |
| API / provider | self-host / partners | self-host / partners | — |
| Modalities | text → text | text → text | — |
| Specs | |||
| Context window | 131K | 131K | tie |
| Max output | 33K | 16K | A +16K |
| Parameters | 117B | 21B | A +96 |
| Pricing | |||
| Input $/1M | $0.15 | $0.05 | B +$0.10 |
| Output $/1M | $0.60 | $0.20 | B +$0.40 |
| Blended $/1M (3∶1) | $0.26 | $0.09 | B +$0.17 |
| Speed | |||
| tok/s | 70 | 120 | B +50 |
| TTFT (s) | 0.4 | 0.2 | B +0.20 |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | 78.2 | 68.8 | A +9.4 pts |
| Humanity's Last Exam | 19.6 | 11.0 | A +8.6 pts |
| AIME 2025 | 92.5 | 91.7 | A +0.8 pts |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 21.8 | 16.0 | A +5.8 pts |
| AA-LCR v1.1 | 52.0 | 34.7 | A +17.3 pts |
| CritPt | 1.1 | 1.4 | B +0.3 pts |
| MMMU-Pro | — | — | — |
| IFBench | 69.0 | 65.1 | A +3.9 pts |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | 26.0 | 60.7 | B +34.7 pts |
| SWE-bench Pro | 16.2 | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | — | — | — |
| Terminal-Bench 2.1 | 26.2 | 13.9 | A +12.3 pts |
| Aider Polyglot | 41.8 | 34.2 | A +7.6 pts |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | 34.0 | 38.9 | B +4.9 pts |
| CursorBench | — | — | — |
| SWE-Rebench | 27.3 | 8.1 | A +19.2 pts |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | — | — |
| WebDev Arena | — | — | — |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | 1,352 | 1,317 | A +35 Elo |
gpt-oss-120B
- Context
- 131K / 33K out
- Parameters
- 117B
- Price
- $0.15 / $0.60
- Speed
- 70 tok/s · 0.4s TTFT
- Modalities
- text → text
- License
- Apache 2.0
OpenAI's open-weight 120B release under Apache 2.0 for self-hosted frontier-adjacent chat.
gpt-oss-20B
- Context
- 131K / 16K out
- Parameters
- 21B
- Price
- $0.05 / $0.20
- Speed
- 120 tok/s · 0.2s TTFT
- Modalities
- text → text
- License
- Apache 2.0
Compact open gpt-oss variant that runs on a single consumer GPU.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.