Inception Mercury 3
Mercury 3 is Inception
Labs' flagship diffusion language model — optimized for low-latency chat,
agents, and voice-adjacent workloads (1M-token context, up to 131K output tokens,
1000+ tokens/s on NVIDIA GPUs). OpenTryOn exposes the hosted chat path on
understand (text in, text out).
| Registry id | mercury-3 |
| Adapter | InceptionUnderstandAdapter |
| Auth | INCEPTION_API_KEY |
| Default base URL | https://api.inceptionlabs.ai/v1 |
| Upstream model id | mercury-3 (also --inception-model mercury-2.5, mercury-2, mercury-voice) |
Mercury 3 on v1/chat/completions is text-only (no --image). For
classification and scoring with optional images, use
Mercury Decide on decide --model mercury-decide.
CLI
export INCEPTION_API_KEY="..."
opentryon understand --model mercury-3 \
--prompt "Draft a 2-sentence product description for a merino crewneck." \
--reasoning-effort medium --dry-run
--reasoning-effort accepts instant, low, medium, or high.
--temperature must stay between 0.5 and 1.0 when set.
Python
from tryon.api.inception import InceptionUnderstandAdapter
adapter = InceptionUnderstandAdapter() # INCEPTION_API_KEY
result = adapter.understand(
prompt="Summarize this merchandising brief in three bullets.",
reasoning_effort="medium",
max_tokens=2048,
)
print(result["text"])
MCP
Tool: understand_mercury_3. Restart MCP so TryOn Studio lists it. Connect
saves INCEPTION_API_KEY.