Blog
Product updates, model launches, and integration guides.
GLM-6: Release Date Predictions and What to Expect vs GLM-5.3
GLM-6 hasn't been announced yet. Here's what Zhipu has actually said about it, when it's most likely to ship based on the GLM release history, and how it could compare with GLM-5.3.
DeepInfer and DeepSeek Recipe: Open Models, Open Kernels, Open Inference
DeepSeek open-sourced DeepInfer, the inference framework behind DeepSeek-V4.1-Flash, along with deepseek-recipe, the encoding layer behind its official API. Here is what each repo contains and why the Huawei Ascend support caught our attention.
MiniMax M3.1 Flash: Frontier Agent Scores at Flash Prices, Now on A2Agent
MiniMax released M3.1 Flash on October 19: top scores on Automation-Bench and LOCA-Bench, 200+ tokens per second, and $0.15 / $0.60 per million tokens. Benchmarks, cost per task, and how to call it on A2Agent at half the official price.
DeepSeek V4.1 Flash 1017 Is Live on A2Agent: Same Price, Stronger Agent Scores
DeepSeek released V4.1-Flash-1017 on October 17, an updated V4.1 Flash checkpoint that improves on all 11 published benchmarks at the same API price. It's now live on A2Agent. Benchmarks, cost per task, pricing, what changed under the hood, and how to call it.
Welcome to A2Agent: The Cheapest AI Model API for Open-Source Models
A2Agent is the cheapest AI model API for the top open-source model families: 22 models from DeepSeek, Qwen, GLM, Kimi and MiniMax behind one key, priced up to 50% below the vendors' own rates. What you get, how pricing works, and how to connect your code and coding tools.
InkTrace: Using the Jev Model to Detect AI-Written Articles in Seconds
InkTrace is an AI-powered Chrome extension that calls TypeSafe's Jev model to tell you, in seconds, whether an article was written by AI or by a person. Here's what it does and how to use it.
REA Deep Dive: How AI Agents Reverse Engineer Software
A look at REA (Reverse Engineer Anything), an open-source project that connects AI agents to software analysis tools through MCP, with a walkthrough of its Notion clipboard case study and how to get started.
A Deep Dive into the Traits of Seven Open-Weight Models
Seven open-weight LLMs compared (DeepSeek V4.1 Flash, Qwen 3.8-Max, GLM-5.3-Flash, Kimi K3, MiniMax M3, Gemma 4 and Mistral Large 4): scale, architecture, licenses, cost per task, self-hosting estimates and picks by use case, all traceable to official pages as of October 2026.
Free AI APIs in 2026: Quotas, Catches and Who Each One Fits
A field guide to nine free AI API options in 2026 (Gemini, Groq, SambaNova, Cerebras, OpenRouter, Puter, LM Studio, Zhipu and Alibaba Model Studio) covering each one's free quota, catches, setup steps and the developers it suits, plus our open GitHub directory of free model, search, embedding, OCR and speech APIs.
Cheapest AI Models for Code in 2026: Price Comparison on A2Agent
All 21 priced models on A2Agent ranked by cost for coding work: DeepSeek, GLM, Kimi, Qwen and MiniMax, with official vs. A2Agent prices and monthly totals for three coding workloads.
Xiaomi MiMo-V2.6: Training in Public, and Why the Model Race Is Still Open
Xiaomi's MiMo-V2.6 Pro and Flash are more than another open-weight release: an open training process, a big generational jump, and a sign that the model landscape, and the opportunity built on it, is far from settled.
Closed vs. Open-Source LLMs: Current State, Capability Gap, and How to Choose
Using Arena and Vals Index as a common yardstick, this post measures how far open-source LLMs trail closed models on user preference and real-task completion, and how cost should factor into choosing between them.
Picking an Agent Client After Ten Years in Engineering: Codex, OpenClaw, and AgentiLoop
Why Codex hides too much and OpenClaw stalls on deep tool chains, and what AgentiLoop/Agent does differently — AgentScript, privilege separation, Accessibility-based automation, and the harness details that decide whether a long task finishes.
From Handwritten Code to the Malleable Computer: DHH's Vision for the Agentic Era
After reading Lex Fridman's long conversation with DHH, we ask a simple question: as AI gets better at building software, what will become truly scarce?
Your LLM Benchmark Is Probably Measuring the Wrong Reality
The AI Observatory puts 24,521 real conversations under one taxonomy and finds that AI usage changes with the dataset, the model, and the month. For anyone shipping an LLM product, that turns "which model is best?" into a much more interesting systems problem.
Kimi-K3's Frontend Win Complicates the U.S.-China AI Narrative
Kimi-K3 tops the Frontend Code Arena at 1,679, briefly flipping the U.S.-China gap. Here is what the chart does — and does not — prove about the AI race.
How to Integrate A2Agent with the OpenAI SDK
Already using the OpenAI SDK? Switching to A2Agent takes two lines — set the base URL and your A2Agent key.
New Models Added This Month
We continually expand the catalog. Here are the latest additions and what they are good at.