DedeProGames/Astral-2.7:20b

DedeProGames/ Astral-2.7:20b

1 Download Updated 14 hours ago

Astral-2.7 is a 20B-parameter AI model based on gpt-oss:20b, designed for high performance on general reasoning tasks.

tools thinking 20b

ollama run DedeProGames/Astral-2.7:20b

curl http://localhost:11434/api/chat \
  -d '{
    "model": "DedeProGames/Astral-2.7:20b",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='DedeProGames/Astral-2.7:20b',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'DedeProGames/Astral-2.7:20b',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 14 hours ago

14 hours ago

8d70a8839168 · 14GB ·

model

archgptoss

·

parameters20.9B

·

quantizationMXFP4

14GB

template

<|start|>system<|message|>You are Astral, a large language model trained by DedeProGames. Knowledge

7.4kB

license

Could not load data.

0B

params

{ "temperature": 1 }

18B

Readme

Astral-2.7 is a 20B-parameter AI model based on gpt-oss:20b, designed for high performance on general reasoning tasks.

20B parameters for strong capability and stability
Built on top of gpt-oss:20b
Optimized for high general reasoning performance
Aims to deliver better reasoning performance than the base gpt-oss:20b
Two variants available: a standard model and an extra-high reasoning version

Benchmark	Métrica	gpt-oss:20b	Astral-2.7
AIME 2024 (no tools)	accuracy %	80.0	92.1
AIME 2024 (with tools)	accuracy %	86.0	96.0
AIME 2025 (no tools)	accuracy %	72.1	91.7
AIME 2025 (with tools)	accuracy %	90.4	98.7
GPQA Diamond (no tools)	accuracy %	66.0	71.5
GPQA Diamond (with tools)	accuracy %	67.1	74.2
HLE (no tools)	score %	7.0	10.9
HLE (with tools)	score %	8.8	17.3
MMLU	accuracy %	84.0	85.3
MMMLU (Average)	accuracy %	73.5	75.7
SWE-Bench Verified	pass@1 %	53.2	60.7
Codeforces (no tools)	Elo	1998	2230
Codeforces (with tools)	Elo	2064	2516
Aider Polyglot	pass@1 %	26.6	34.2
Tau-Bench Retail	score %	47.3	54.8
Tau-Bench Airline	score %	42.6	38.0
HealthBench	score %	41.8	42.5
HealthBench Hard	score %	12.9	10.8
HealthBench Consensus	score %	83.0	82.6

Install

Install the Standard Model

ollama pull DedeProGames/Astral-2.7:20b

Install the xhigh reasoning Model

ollama pull DedeProGames/Astral-2.7:20b-plus

![ChatGPT Image 3 de mar. de 2026, 18_01_40.png](/assets/DedeProGames/Astral-2.7/a218b0c6-7a8c-4048-8fd8-b565c00b1188)

Astral-2.7 is a **20B-parameter AI model** based on **gpt-oss:20b**, designed for **high performance on general reasoning tasks**.

- **20B parameters** for strong capability and stability
- Built on top of **gpt-oss:20b**
- Optimized for **high general reasoning performance**
- Aims to deliver **better reasoning performance than the base gpt-oss:20b**
- Two variants available: a standard model and an extra-high reasoning version

---

| Benchmark | Métrica | gpt-oss:20b | Astral-2.7 |
|---|---:|---:|---:|
| AIME 2024 (no tools) | accuracy % | 80.0 | 92.1 |
| AIME 2024 (with tools) | accuracy % | 86.0 | 96.0 |
| AIME 2025 (no tools) | accuracy % | 72.1 | 91.7 |
| AIME 2025 (with tools) | accuracy % | 90.4 | 98.7 |
| GPQA Diamond (no tools) | accuracy % | 66.0 | 71.5 |
| GPQA Diamond (with tools) | accuracy % | 67.1 | 74.2 |
| HLE (no tools) | score % | 7.0 | 10.9 |
| HLE (with tools) | score % | 8.8 | 17.3 |
| MMLU | accuracy % | 84.0 | 85.3 |
| MMMLU (Average) | accuracy % | 73.5 | 75.7 |
| SWE-Bench Verified | pass@1 % | 53.2 | 60.7 |
| Codeforces (no tools) | Elo | 1998 | 2230 |
| Codeforces (with tools) | Elo | 2064 | 2516 |
| Aider Polyglot | pass@1 % | 26.6 | 34.2 |
| Tau-Bench Retail | score % | 47.3 | 54.8 |
| Tau-Bench Airline | score % | 42.6 | 38.0 |
| HealthBench | score % | 41.8 | 42.5 |
| HealthBench Hard | score % | 12.9 | 10.8 |
| HealthBench Consensus | score % | 83.0 | 82.6 |

---

## Install

### Install the Standard Model
```bash
ollama pull DedeProGames/Astral-2.7:20b
```

### Install the xhigh reasoning Model
```bash
ollama pull DedeProGames/Astral-2.7:20b-plus
```

Paste, drop or click to upload images (.png, .jpeg, .jpg, .svg, .gif)