Supported Models
Reference of all models in the emissions.dev LLM database with their energy tier and reasoning configuration.
<div class="admonition admonition-warning"><span class="admonition-icon">⚠️</span><div class="admonition-content"><p>Beta
Model coverage is updated regularly. If your model isn't listed, the API estimates its tier from the model name. Pass any model string — unknown models return a <code>unknown_model</code> notice with the estimated tier.</p>
</div></div>
OpenAI
GPT-5.x
| Model |
Tier |
Reasoning |
gpt-5.2 |
frontier |
— |
gpt-5.2-pro |
frontier |
5× thinking ratio |
gpt-5.2-codex |
frontier |
— |
gpt-5.1 |
frontier |
— |
gpt-5.1-codex |
frontier |
— |
gpt-5 |
frontier |
— |
gpt-5-pro |
frontier |
5× thinking ratio |
gpt-5-mini |
medium |
— |
gpt-5-nano |
small |
— |
GPT-4.x
| Model |
Tier |
Reasoning |
gpt-4.5 |
frontier |
— |
gpt-4.1 |
large |
— |
gpt-4.1-mini |
medium |
— |
gpt-4o |
large |
— |
gpt-4o-mini |
medium |
— |
gpt-4-turbo |
large |
— |
gpt-4 |
large |
— |
gpt-3.5-turbo |
medium |
— |
Reasoning (o-series)
| Model |
Tier |
Thinking Ratio |
o3 |
frontier |
6× |
o3-mini |
large |
4× |
o3-deep-research |
frontier |
15× |
o4-mini |
large |
3.5× |
o4-mini-deep-research |
large |
10× |
o1 |
frontier |
5× |
o1-mini |
large |
3× |
o1-pro |
frontier |
8× |
Open Source
| Model |
Tier |
Reasoning |
gpt-oss-120b |
frontier |
— |
gpt-oss-20b |
medium |
— |
Anthropic
Claude 4.5
| Model |
Tier |
Reasoning |
claude-opus-4.5 |
frontier |
— |
claude-sonnet-4.5 |
frontier |
— |
claude-haiku-4.5 |
medium |
— |
Claude 4.x
| Model |
Tier |
Reasoning |
claude-opus-4.1 |
frontier |
— |
claude-opus-4 |
frontier |
— |
claude-sonnet-4 |
large |
— |
Claude 3.x
| Model |
Tier |
Reasoning |
claude-3-opus |
frontier |
— |
claude-3.7-sonnet |
large |
— |
claude-3.5-sonnet |
large |
— |
claude-3-sonnet |
large |
— |
claude-3.5-haiku |
medium |
— |
claude-3-haiku |
small |
— |
Google
Gemini 3
| Model |
Tier |
Thinking Ratio |
gemini-3-pro |
frontier |
— |
gemini-3-flash |
large |
— |
gemini-3-deep-think |
frontier |
6× |
Gemini 2.5
| Model |
Tier |
Thinking Ratio |
gemini-2.5-pro |
frontier |
— |
gemini-2.5-pro-deep-think |
frontier |
6× |
gemini-2.5-flash |
large |
— |
gemini-2.5-flash-lite |
medium |
— |
Gemini 2.0 / 1.5
| Model |
Tier |
Reasoning |
gemini-2.0-pro |
large |
— |
gemini-2.0-flash |
medium |
— |
gemini-1.5-pro |
large |
— |
gemini-1.5-flash |
medium |
— |
gemini-ultra |
frontier |
— |
gemini-pro |
large |
— |
gemini-nano |
small |
— |
Llama 4
| Model |
Tier |
Reasoning |
llama-4-behemoth |
frontier |
— |
llama-4-maverick |
large |
— |
llama-4-scout |
medium |
— |
Llama 3.x
| Model |
Tier |
Reasoning |
llama-3.3-70b |
large |
— |
llama-3.2-90b |
large |
— |
llama-3.2-11b |
medium |
— |
llama-3.2-3b |
small |
— |
llama-3.1-405b |
frontier |
— |
llama-3.1-70b |
large |
— |
llama-3.1-8b |
small |
— |
llama-3-70b |
large |
— |
llama-3-8b |
small |
— |
Mistral
| Model |
Tier |
Reasoning |
mistral-large-2 |
large |
— |
mistral-large |
large |
— |
mistral-medium |
medium |
— |
mistral-small |
small |
— |
mixtral-8x22b |
large |
— |
mixtral-8x7b |
medium |
— |
DeepSeek
| Model |
Tier |
Thinking Ratio |
deepseek-v3 |
large |
— |
deepseek-v3.1 |
large |
— |
deepseek-r1 |
frontier |
6× |
deepseek-r1-lite |
large |
3× |
xAI
| Model |
Tier |
Reasoning |
grok-4 |
frontier |
— |
grok-4.1 |
frontier |
— |
grok-3 |
frontier |
— |
grok-2 |
large |
— |
Cohere
| Model |
Tier |
Reasoning |
command-r-plus |
large |
— |
command-r |
medium |
— |
Self-Hosted
Generic tiers for self-hosted deployments. Use these when your model isn't in the database:
| Model |
Tier |
Reasoning |
self-hosted-large |
large |
— |
self-hosted-medium |
medium |
— |
self-hosted-small |
small |
— |
<div class="admonition admonition-info"><span class="admonition-icon">ℹ️</span><div class="admonition-content"><p>Self-hosted models use a 1.5× efficiency multiplier to account for higher PUE and less optimised hardware compared to hyperscale providers. Pass <code>region</code> to use your actual grid intensity.</p>
</div></div>
Unknown Models
Any model string not in this list is automatically classified using pattern matching:
| Pattern |
Estimated Tier |
Contains gpt-5, opus, ultra, behemoth, 405b, o3, grok-3, grok-4 |
frontier |
Contains gpt-4, 70b, sonnet, pro, maverick, large, r1 |
large |
Contains gpt-3.5, mini, scout, flash, medium |
medium |
Contains nano, tiny, small, 7b, 8b, haiku, lite |
small |
| No match |
medium (default) |
The response will include a unknown_model notice with the estimated tier. Check inference.is_known_model to detect this programmatically.
Energy Factors by Tier
| Tier |
Base Energy (Wh/request) |
Input (Wh/1K tokens) |
Output (Wh/1K tokens) |
Embodied (gCO₂e/request) |
| frontier |
0.50 |
0.050 |
0.250 |
0.30 |
| large |
0.25 |
0.020 |
0.100 |
0.15 |
| medium |
0.10 |
0.005 |
0.025 |
0.05 |
| small |
0.03 |
0.001 |
0.005 |
0.01 |