Structured outputs are very underrated. It's often much easier to constrain LLM outputs to a JSON schema than asking for a tool call.

OpenRouter now normalizes structured outputs for
- 46 models
- 8 different model companies
- 8 free models!

Demo:

Llama 3.3 (shown in brown) is now 35% of Llama token traffic 👀
openrouter.ai/meta-llama

PydanticAI now supports OpenRouter!

You can now filter for LLM providers on our /models page

OpenRouter has prevented 13k Sonnet requests from failing in the last 2 days

LLMs are getting insanely cheap – just analyzed 500+ websites for under $0.01 via @openrouter.bsky.social using llama3.1 🤯

First post here, and we're excited to announce that Discourse AI now supports @openrouter.bsky.social!

With hundreds of models, including free options, AI is now more accessible. Select your provider and fine-tune results in the Discourse UI.

Learn more 👇
meta.discourse.org/t/openrouter...

ICYMI, lots of price cuts last week. helping you get best-in-market:

🦙Llama 3.2 3B Instruct: 40% cheaper, now $0.018 (was $0.03)
🦙Llama 3.1 Nemotron: $0.20 (was $0.35)
🦙Llama 3.1 8B: $0.30 (was $0.60)
🦙Llama 3.1 70B: $0.20 (was $0.40)
🦙Llama 3.1 405B: $0.90 (was $2.80)

Amazon Nova is now live on OpenRouter ☄️
...along with Bedrock! more coming soon

Only took 40 minutes

Llama 3.3 is live! 🦙🦙🦙
openrouter.ai/meta-llama/l...

OpenRouter now generates about 'one Wikipedia' worth of words every day

Overall $ volume on OpenRouter is up by over 2300% 📈
even though LLM prices declined by 90% 📉
due to a lotta tokens:

The Reality for AI Startups

QwQ 32B, an open-source model comparable with o1-preview and Claude 3.5 Sonnet, is now live!

Scores 50 on the AIME, compared to
- 44.6 for o1-preview
- 16 for Sonnet

Try it out here: openrouter.ai/qwen/qwq-32b...