Fastest inference anywhere: open models on-prem, hosted or on-device. Backed by @ycombinator W26
Joined July 2025
- Tweets527
- Following6
- Followers2.5K
- Likes1.2K
Pinned Tweet
Meet Wally.
Our inference stack for open frontier models, built to be the fastest place to run them.
Performance snapshot:
GLM-5.3 Flash: 380 tok/s.
GLM-5.3 Max: 790 tok/s
Qwen3.8-27B: 485 tok/s
DeepSeek-V4.1 Flash: 615 tok/s
1/7
RunAnywhere retweeted
Steam reviews are a war zone. I’m not reading all that 📷
So we put DeepSeek V4.1 Flash on Wally and asked it what players were actually complaining about.
Less scrolling. More answers.
RunAnywhere retweeted
🚨A new tier of frontier models dropping VERY soon.🚨
And yes, we're launching a special FAST mode for those who want warp speed. ⚡
We just got a LOT more compute. Stay Tuned!
RunAnywhere retweeted
You think the open models you use right now are fast?
Try them on Wally.
Sketch an app in tldraw, hit Make Real, and DeepSeek V4.1 Flash has it running before you let go of the mouse. Crazy fast.
Agreed DeepSeek-v4.1-flash is an amazing model.
and we fixed the speed, it runs at 615+ tok/s on Wally
here it is against DeepSeek's own api making changes to the 2048 game -:
RunAnywhere retweeted
Letss go! DeepSeek V4.1 Flash-1017 is the best open model for web dev, and it runs at 615 tok/s on Wally(@RunAnywhereAI).
we gave it the live Hacker News front page and it redesigned it as if Apple made it, then as a 1999 GeoCities page and it was built instantly.
Exciting news: DeepSeek-V4.1-Flash-20261017 (Max) by @deepseek_ai is now the #1 open model in Code Arena: WebDev, surpassing Kimi K3 (Max) at 10x lower cost!
Highlights:
- #1 open model, ahead of Kimi K3 (Max)
- #12 overall with 1670 pts, ahead of Claude Opus 5 (High) and within ~10 pts of GPT-6 Sol (Max)
- Shifts the Pareto frontier at $0.3/$1.2 per million input/output tokens
- +52 pts vs. the previous DeepSeek-V4.1-Flash-20260910
Congrats to the @deepseek_ai team on another strong contribution to the open source ecosystem!
DeepSeek V4.1 Flash on Wally is now upgraded to 1017.
- 77% on DeepSWE, up from 74.2%
- ~$0.20 per task, down from ~$0.50
- Shorter outputs, so tasks finish sooner
At 615 tok/s. Same model id, nothing to change on your side.
🚀 DeepSeek-V4.1-Flash-1017 is here!
Stronger benchmark performance, faster task completion, and shorter outputs than the previous version—a Pareto improvement in performance and cost. Now live on our website, app, and API.
🛠️ Optimized for @DeepSeekHarness, Tencent CodeBuddy, WorkBuddy, and more. Upgrade to DeepSeek Harness v0.2.1-rc.1 or later for the best experience with the new model.
1/4
RunAnywhere retweeted
Everyone's shipping speed now.
Model quality is hitting diminishing returns. A 3-point benchmark lead is a vanity metric you can't feel in real work.
Speed you feel on every prompt. Might as well run models where they're actually the fastest 🏎️ @RunAnywhereAI
EVE just got an upgrade.
Now on @perplexity_ai's new open pplx-decider-v1.2:
-Smarter at every kind of decision
-150 ms → 100 ms per decision
-Flash v1.2 runs on your laptop with EVE local
$5 free on signup.
RunAnywhere retweeted
@AIatMeta's Muse Spark Open just landed on Wally, the best US open model, and it's running at 370+ tok/s. Faster than any other provider out there.
So obviously the first thing I did was ask it for a Snake game. One prompt, under 15 seconds, and it was already playing itself 🐍
Muse Spark Open is now live on Wally, day 0
Best US open model thanks to @finkd
370 tok/s, ~2x faster than any provider on OpenRouter.
$5 Free on signup.
RunAnywhere retweeted
The #1 US open-weights model, and it's already live on Wally(@RunAnywhereAI) at 370 tok/s.
For reference, the fastest provider @ArtificialAnlys lists for @AIatMeta's Muse Spark Open right now is at 216.
$5 free on signup, so try it right now at the fastest speeds possible!
Muse Spark Open is now live on Wally, day 0
Best US open model thanks to @finkd
370 tok/s, ~2x faster than any provider on OpenRouter.
$5 Free on signup.
The OpenRouter board for Muse Spark Open right now, P50 output speed:
@baseten: 196 tok/s
@FireworksAI_HQ : 61 tok/s
@nebiusai : 58 tok/s
@togethercompute : 32 tok/s
Wally: 370 tok/s. On an 816B model.
Copy this into your agent to get started:
"set up runanywhere.ai/SKILL.md"
Or straight from the terminal:
wally opencode -m muse-spark
runwally.com
RunAnywhere retweeted
Quantum entanglement on Wikipedia is unreadable, so I made the page explain itself to me like I'm 5.
Every paragraph rewrites itself the second I scroll to it. Halfway down I got bored and asked for a sports commentator instead. Einstein vs Bohr has never been this hype.
DeepSeek V4.1 Flash on Wally. It rewrote the paragraph faster than I could blink at it.