Sansa

Couldn't agree more @DavidSacks Sansa is that middleware: - routing in under 20ms - outperforms frontier - less than half the cost We are lowering the cost of intelligence

Leaking the MMLU-Pro results for sansa-auto (live tomorrow). We just beat Claude Mythos Preview at 1/280th the cost. sansa-auto (routed): - 93.11% - $9.05 per benchmark run mythos preview: - 92.70% - $2,531 per benchmark run Cost/performance is the new meta.

Our Unified Gateway is live! + 200 plus models, one API trysansa.com/product/unifie…

Today we are launching our unified gateway product - call 200+ models from 30+ labs - lower price point than openrouter - openai compatible API AND sansa-auto is now transparent: - see exactly what model was called - only pay for what you need We want to build the go to API for building AI products What would you like to see from Sansa?

We had an authentication outage overnight. Everything is back up and fully operational as of 3:32 AM PT. The issue effected our API and web application. We caught the issue early and have already implemented a fix. If you were impacted (unexpected auth error), please reach out to @jbm_dev or @sansa_ai so we can make it right!

Resolved

Performance Degraded: API latency has increased over the last 10min, working on a fix!

SONNET 4.6 BENCH RESULTS We ran Sonnet 4.6 through our benchmark. Non-reasoning beats GPT-5.2 with reasoning. Let that sink in.

Our launch has been rescheduled! Launching: Monday Feb 16th 2026 Silver lining: 2 extra weeks to make sure you are on the waitlist so you get free credits on launch day

Coming 01.31.2026

11.7 Billion!

Sansa AI Inc. is not affiliated with or endorsing any cryptocurrencies, including any unsolicited tokens using our name/IP.

Have you ever picked a worse AI model for your product because it was cheaper?

6 days later: - 227 waitlist signups - 29k+ page visits - 1600 followers - 3.1M impressions - clarity of vision build in public. don't give up.

@xeno_nft199 Of course! you can reach out directly to @Jbm_dev for faster response

@Sansa_router Hey, had a couple of questions about the project and wanted to learn more Would it be okay to reach out in DMs?

Newest Sansa Benchmark data is out! - OpenAI's models are the most censored - GPT-5-mini beat GPT-5.2 on agentic tasks - Grok 4.1 Fast is the best value by far - Gemini 3 Pro took the top spot overall What's new 👇

@MazenALDamshity We think this is mostly due to efficiency, the smaller model explores less and wastes less time

@Sansa_router 5 mini outperforming 5.2 is wild ngl 🤣🤣🤣

Data & Methodology: trysansa.com/benchmark

on em dash resistance: - all models love the em dash - Qwen models are the worst offenders - parameter count doesn't predict this behavior - reasoning mode doesn't predict it either - most models land in the same range not much to conclude here except they're all addicted