Do you always need a frontier proprietary model for coding agents? 89.6% at $0.02 per task vs 87.0% at $1.00. That's GLM-5.3-Flash next to GPT-6 Astra on SWE-bench Verified. The cheaper model solved more tasks. We ran 12 configs of 9 models on accuracy, cost, and time. Where open models win and where they don't, check it out 👉 friendli.ai/blog/open-vs-clo…

AI agents are taking on increasingly ambitious tasks, even Millennium Prize Problems 🫢. As tasks get longer and more complex, cost per token tells less of the story. Cost per completed task starts to matter much more. Cheaper tokens ≠ cheaper tasks. GLM-5.3 and GLM-5.3-Flash both beat GLM-5.2 on accuracy at a lower cost per task. Flash gets there with cheap tokens. GLM-5.3 gets there with 12–45% fewer agent turns, at the exact same price as 5.2. Read the full blog 🔗 friendli.ai/blog/glm-5.3-vs-…

Switching your coding agent to a new inference provider shouldn't mean digging through JSON, TOML, and YAML files. Introducing FriendliLink 🔗: An open-source CLI (frlink) that points the coding agent you already use — Claude Code, Codex, Cursor, OpenCode, Pi, Hermes Agent, DeepSeek Harness — at open models on FriendliAI Model APIs. One command to connect, one command to put it back. What it offers: → No config hunting — frlink claude on writes the key and base URL into the agent's own settings → Fully reversible — off restores your original config byte-for-byte → Reasoning that works — thinking controls pulled from Friendli's live catalog, translated to what each agent expects → No proxy, no daemon — after on, your agent talks to Friendli directly 👉 Get your API key on friendli.ai/suite 👉 Full blog: friendli.ai/blog/friendlilin…

FriendliAI is now a BYOK provider on @vercel AI Gateway. 🎉 Already on the Gateway? Route to FriendliAI with no SDK swap — just bring your Friendli API key. Pricing and credits carry through. ✨ GLM-5.2 and Gemma 4 listed today. Get a Friendli API key: friendli.ai/docs/guides/suit…

🚀 Day-0 support for GLM-5.3 is live on Friendli Model APIs. @Zai_org ’s latest open-weight model delivers major gains in coding and long-horizon agentic tasks, including open-weight SOTA on Terminal Bench 3.0 and Agents’ Last Exam. 🔗 Try GLM-5.3 now friendli.ai/models/zai-org/G…

Our team put two of the latest frontier open-weight models, GLM-5.3 and Kimi K3, to the test.

GLM-5.3 vs. Kimi K3: Which coding agent actually delivers? 🤔 We evaluated both models across standard coding agent benchmarks and an open-ended game development task. Read the full breakdown 👉 friendli.ai/blog/kimi-k3-glm… Get on the GLM-5.3 and Kimi K3 drop list 👉 qg2m6.share.hsforms.com/2qG7…

Interviewing our wonderful panel from last night! Thank you to all who showed up and listened in. We enjoyed hosting @nvidia, @OpenHandsDev, and @ollama at our SF office. Stay tuned for the next one! 👀

Your response_format validates. The output is still useless. 🤷 Constrained decoding only masks invalid tokens — it doesn't pick the right one. Structure ≠ intent. Bonus: up to 21% faster on FriendliAI via speculative decoding. Full breakdown 👇

🎂 Stack your AI coding cake! Inspired by @JensenHuang's “AI is a five-layer cake” framework, this game lets you build the full AI coding stack with four essential layers: 🧠 AI Agents 🤖 Open-weight Models 🔵 @friendliai — Inference 🟢 @nvidia — GPUs Inference is where it all comes together—and what FriendliAI does best. Play the game, screenshot your result, and share your AI coding cake in the comments. 👇 stack-your-ai-coding-cake.fr…

Looking forward to the event!

This Thursday, August 20th, we’re bringing together four leaders at the forefront of open models, coding agents, and AI infrastructure: Kyle Kranen, Senior Manager, Dynamo at @nvidia Yunmo Koo, Founding Engineer at @friendliai Dong Chen, Software Engineer at @ollama Saurya Velagapudi, Principal Engineer at @OpenHandsDev Together, they’ll explore what it takes to move AI coding agents to production and how open models are changing the economics of AI-native engineering. Join us for the conversation, followed by audience Q&A and networking. Spots limited! 📍 San Francisco 📅 Aug. 20 RSVP: luma.com/o9tv62y5

GLM-5.3 is out. Top-tier coding and agentic performance!

Can't wait. @friendliai @OpenHandsDev @ollama

Open Models, Production Agents: The New Economics of AI-Native Engineering August 20th, @friendliai is hosting @OpenHandsDev and @ollama for an event full of bringing together leading voices at the intersection of open models, coding agent infrastructure, and agentic engineering. Featuring Saurya Velagapudi, Principle Engineer at @OpenHandsDev and Jeffrey Morgan, CEO at @ollama Spots are limited, grab yours while they’re available! luma.com/o9tv62y5

MiniMax H3: Omni-Reference, Commercial-Grade Generation, Unbeatable Cost Efficiency, Open Weights Your creative destiny, on your terms. Now Live at HailuoAI.video & MiniMax API.

@friendliai has signed the open letter “Open Weights and American AI Leadership,” alongside @Microsoft, @nvidia, @Meta, @OpenAI, @ollama, and 50+ other organizations. Open weight models are the foundation of an accessible, competitive AI ecosystem. They let any team — startup, enterprise, university, or public institution inspect, adapt, and run advanced AI on their own terms, and they keep the benefits of AI broadly shared rather than concentrated in a few hands. We believe openness is one of the most important paths to innovation, competition, and even safety in AI. Glad to stand with this community. Read the letter: microsoft.com/en-us/corporat… #OpenWeights #OpenSourceAI

Welcome to FriendliAI, Brian Cho! 🎉 Joining as Director of Engineering with 6 years at @Meta and 4 at @AptosLabs under his belt — right as our engineering team scales to match the pace of the agent era.🚀

NVIDIA Nemotron 3 Embed is now on FriendliAI. Open embedding models built for agentic retrieval: → 8B: frontier accuracy, #1 on the RTEB leaderboard → 1B: designed to retain 95%+ of the 8B's accuracy, built for scale → 16K context, text + code, fully open weights, datasets, recipes, and fine-tuning guidance Deploy via Friendli Dedicated Endpoints, OpenAI-compatible API. 👉friendli.ai/models/nvidia/Ne… 👉friendli.ai/models/nvidia/Ne… 👉friendli.ai/models/nvidia/Ne… Read full blog 🔗friendli.ai/blog/nvidia-nemo…

Introducing Kimi K3: Open Frontier Intelligence 🔹 2.8 Trillion Parameters, 1 Million Context, Native Multimodal 🔹 Kimi Delta Attention enables up to 6.3x faster decoding in million-token contexts 🔹 Attention Residuals deliver ~25% higher training efficiency at <2% additional cost 🔹 Built for long-horizon agentic coding and self-evolving workflows Kimi K3 is now live on on Kimi.com, Kimi Work, Kimi Code, and the Kimi API. Open Weights by July 27, 2026. 🔗 API: platform.kimi.ai 🔗 Tech blog: kimi.com/blog/kimi-k3