Sabitlenmiş Tweet
Ethan
738 posts
Ethan
@Ethan1cc
Following the AI stack from models to inference. Thoughts on Kimi, DeepSeek, agents, benchmarks, and what's actually useful.
Santa Calra, California Katılım Ağustos 2025
2.4K Takip Edilen2K Takipçiler
Come join us at Computex 2026 this week!
Canopy Wave@CanopyWave_AI
Heading to Computex 2026? Our CTO, James Liao, is on-site this week and available to meet with AI companies, startups, and enterprise teams. If you're exploring AI infrastructure, inference at scale, enterprise AI deployment, or partnership opportunities, we'd love to connect. DM us or leave a comment to set up a meeting. Looking forward to productive conversations at Computex 2026.
English
Ethan retweetledi
@deepseek_ai and @XiaomiMiMo cutting API prices isn't just a pricing story.
It's an infrastructure story.
The biggest breakthroughs in AI inference right now are no longer just model quality.
They're:
"KV cache"
"memory efficiency"
"long-context"
"throughput"
Model architecture is reducing inference costs faster than hardware improvements, making AI infrastructure even more important.
Because in production, the real differentiator is whether a model can consistently deliver its actual capability under real workloads.
Same model ≠ same performance
Reliability, latency, tool calling, and secure inference infra now matter more than ever.
The next AI competition is not only about who trains the best model.
It's about who serves it best.
English
Introducing Lindy Assistant, the ultimate AI assistant.
It talks with you through iMessage, connects to 100s of apps, helps you with your meetings and emails, and proactively finds ways to save you time all day.
Check out some examples of ways Lindy assistant helps below.
English
Ethan retweetledi
A strong model is only the beginning.
What matters in production is:
- stable inference
- secure infrastructure
- predictable performance
- responsive support
That's why infrastructure matters just as much as the model itself.
Kimi K2.6 on Canopy Wave👇
Canopy Wave@CanopyWave_AI
Enterprise adoption of agentic AI depends on more than just model capability. Teams need infrastructure that is reliable, secure, and scalable. Why choose us? · High-quality, stable inference for production workloads · 24/7 engineering support for enterprise customers · SOC 2-certified AI infrastructure · Reliable performance for long-running agentic systems · Transparent pricing without hidden complexity Reliable. Secure. Production-ready. Kimi K2.6 on Canopy Wave.
English
Ethan retweetledi
Enterprise adoption of agentic AI depends on more than just model capability.
Teams need infrastructure that is reliable, secure, and scalable.
Why choose us?
· High-quality, stable inference for production workloads
· 24/7 engineering support for enterprise customers
· SOC 2-certified AI infrastructure
· Reliable performance for long-running agentic systems
· Transparent pricing without hidden complexity
Reliable. Secure. Production-ready.
Kimi K2.6 on Canopy Wave.
English
@kevincodex Hi Kevin, Artemis from Canopy Wave here.
Noticed you’re using Mimo V2.5 — would you be interested in trying our inference platform as well? We also support Kimi K2.6, DeepSeek V4, GLM 5.1, Minimax M2.5, etc.
Happy to set up a quick test if interested.
English
Ethan retweetledi
Still one of the most important model releases this quarter.
GLM-5.1 is becoming one of the strongest open models for long-horizon coding tasks.
In many coding workflows, it's already outperforming Gemini 3.5 Flash.
Now available on Canopy Wave
Try now👇
Canopy Wave@CanopyWave_AI
GLM-5.1 is now live on Canopy Wave Built by @Zai_org, this powerful open-weight model is engineered for long-horizon agentic tasks. It achieves SOTA on SWE-Bench Pro with 58.4, outperforming GPT-5.4 (57.7), Claude Opus 4.6 (57.3), and Gemini 3.1 Pro (54.2). Now available on Canopy Wave's fast, reliable, and secure inference platform — perfect for: - Production-scale agentic engineering - Long-horizon coding workflows - Stable, high-throughput autonomous execution GLM 5.1 is now available in our Coding Plan Pro🎉
English
Finally done w/ tuning, couple changes you wouldn't have guessed matter, but make the number go up:
- Accepting replacement as string[].
Does "" clear a line or delete it? Should it suffix with "\n" for a simple edit? Now it requires no explanation.
- 5:af => 5#ZY.
Smaller models get confused when they see numbers on both sides, swapping to an alpha-only hex dict solves it.
- No autocorrect.
Makes the tool behaviour unpredictable for the smarter ones.
Hashline now wins in every single comparison!
English
Berrydesk is really great — give it a try, it can significantly boost your productivity and make your work much more efficient.
Chirag Asarpota@ChiragAsarpota
Introducing berrydesk.com - AI support agent trained on your docs, Google Drive, and Notion. Works on Slack, Discord, and a beautiful chat bubble on your website. Powered by @tan_stack start @convex @firecrawl @Cloudflare @Netlify @Sentry @autumnpricing Built for the TanStack Start Hackathon. ZERO lines of code -> PROD in just 17 days...
English
Ethan retweetledi
Canopy Wave is officially SOC 2 certified!
This milestone reflects our commitment to delivering high-quality inference services with enterprise-grade security, reliability, and confidentiality.
SOC 2 certification validates that our systems and internal controls are designed to consistently safeguard customer data and ensure a secure, dependable infrastructure for developers and businesses at scale.
Special thanks to Prescient Security for guiding us through the audit and certification process.
English
@Winrey_team9 Hi Winrey, I am Artemis from Canopy Wave.
We would like to talk about cooperation with you.
Wonder if you have any interest
English
Found someone open-source Mythos...
A 22-year-old open-source developer named Kye Gomez pieced together clues from publicly available academic papers and, using pure PyTorch, managed to reproduce this hypothetical architecture.
OpenMythos takes a completely different approach: instead of stacking more layers, it runs the same set of weights over and over again.
You can think of it like this: traditional models are like reading a book and you turn one page and move on to the next, and when you finish, it’s done.
OpenMythos is like one person repeatedly reading the exact same paragraph, understanding it more deeply with each pass.Inference depth no longer depends on how many parameters you have, but on how many times you’re willing to let the model think.
The result?
A 770M-parameter recurrent model can match a 1.3B-parameter traditional Transformer. Same performance, with almost half the parameters.
Check this out⬇️
github.com/kyegomez/OpenM…
English
Ethan retweetledi
On Canopy Wave, you're not just buying tokens — you're getting premium quality tokens.
Deeper reasoning, reliable tool calling, and consistent excellence even past 100k+ contexts.
High-quality inference isn't optional; it's everything.
Canopy Wave@CanopyWave_AI
Real users. Real feedback. Real performance. Users across X, Reddit, and Discord are sharing the same experience with Canopy Wave: ⚡️Fast speeds 🧠Powerful reasoning with Kimi K2.6 🔒Reliable infrastructure 💬Responsive support Unlimited Token Plan users are loving it. Thank you to everyone building with Canopy Wave. Exciting things ahead.
English
Ethan retweetledi
⬆️Full Model Lineup Upgrade
We’ve rolled out a full upgrade across our model stack:
· Kimi K2.5 → Kimi K2.6
· GLM 5 → GLM 5.1
· DeepSeek V3.2 → DeepSeek V4 Flash
· MiMo V2 Flash → MiMo V2.5
Delivering clear improvements across coding, agent workflows, multimodal tasks, and complex reasoning.
Haven’t tried the new models yet?
Now’s the perfect time👇
English
been doing a ton of work with two plus codex subs, opencode go sub (was 5$ for first month) and a few dollars of openrouter credits (deepseek-v4-flash and qwen3.6-27b are incredibly cheap and reliable models).
main goal since spring of last yr was to get the most out of a 100$ ai budget... by fall that fell to around 50$. one can do alot with the current ensemble of subs i just mentioned, more than enough for most people imo.
English
How much do you spend (personally & professionally) on AI per month?
Subscriptions plus compute.
English
@hibrahimkalkan Hi I am Artemis from Canopy Wave.
We offer open-source LLMs.
Wonder if we could talk about cooperation with ABP studio
English
AI Coding Agent is coming to ABP Studio, making the ABP Platform a player in AI-Driven development.
Join the next ABP Community Talks event to learn what it is and how it works.
👉 kommunity.com/volosoft/event…
English
@aaronjmars Really impressive work — turning a single document into a full multi-agent “synthetic public” with feedback loops between social dynamics and markets is a genuinely novel and powerful idea.
Would love to explore ways we could collaborate on this!
English
Ethan retweetledi
Runs at 1/89 the cost of Claude Opus
1/53 of Sonnet
1/18 of Haiku
DeepSeek V4 Flash is live on Canopy Wave⚡️
$0.14 input | $0.28 output | $0.028 cache
248B MoE · 1M context · Built for agents
When your agents scale, cost becomes the bottleneck.
Optimize early. Scale safely.👇
canopywave.com/models/deepsee…
Canopy Wave@CanopyWave_AI
DeepSeek-V4-Flash is now live on Canopy Wave🎉 A fast, cost-efficient MoE model built for production AI workloads: - 284B total/13B active params - 1M token context - Reasoning close to V4 Pro - Strong on simple agent tasks ⚡️Fast + low latency 🤖Great for agents & coding ⚡️High cost-performance ratio $0.14 input/$0.28 output/$0.028 cache Try it now👇
English
Ethan retweetledi
Agent Capability Showcase — Hermes Agent vs OpenClaw
Powered by our Kimi K2.6, both agents received the identical high-difficulty instruction set:
1. Find & summarize today's top 3 AI news
2. Code + save Fibonacci sequence (first 20 terms)
3. Scrape HN top 5 (titles + comments)
4. Build a full inference-benchmark project folder from scratch
Full side-by-side comparison video is out now
Watch the Unlimited Agent showcase!
Which one do you think won? Drop your pick in comments👀
English
Ethan retweetledi
Real users. Real feedback. Real performance.
Users across X, Reddit, and Discord are sharing the same experience with Canopy Wave:
⚡️Fast speeds
🧠Powerful reasoning with Kimi K2.6
🔒Reliable infrastructure
💬Responsive support
Unlimited Token Plan users are loving it.
Thank you to everyone building with Canopy Wave. Exciting things ahead.
English