testified.ai Logo

Qwen 3.5, Dreamer Platform Launch, and New OpenAI Codex Models

Today's AI tooling landscape is defined by major model releases and platform launches. Qwen 3.5 has arrived as a massive open-weight challenger from Alibaba, while Dreamer introduces a new paradigm for building software through conversation. OpenAI continues to iterate with faster coding models and security features, and new agentic tools like Manus and Lemon are redefining user interaction.

Alibaba Releases Qwen 3.5 Series

Alibaba has officially released the Qwen 3.5 series, headlined by the Qwen3.5-397B-A17B model. This vision-language model utilizes a sparse Mixture-of-Experts (MoE) architecture, activating only 17 billion parameters per forward pass despite its 397 billion parameter total. It demonstrates frontier-level performance in reasoning, coding, and agent capabilities, reportedly rivaling GPT-5.2 and Claude Opus 4.5. The model supports 201 languages and is designed to be cost-effective, with Alibaba claiming it is 60% cheaper to use than its predecessor. You can read the full release details here.

Dreamer: Building Apps by Talking

A new platform called Dreamer has launched, founded by former Stripe CTO David Singleton and Hugo Barra. Dreamer allows users to build agentic applications simply by describing them to an AI agent named "Sidekick." The platform handles all infrastructure, hosting, and deployment, removing technical hurdles for creators. It has moved to public beta with a partnership with Anthropic.

OpenAI Updates: Codex Spark and Lockdown Mode

OpenAI has released GPT-5.3-Codex Spark, a smaller, faster model designed for speed. It is reportedly 3x-5x faster than the standard Codex model, features a 128k context window, and runs on Cerebras hardware. Concurrently, OpenAI introduced a Lockdown Mode for ChatGPT, which disables certain features to prevent data exfiltration via prompt injection, catering to security-conscious users. More details on the security update can be found here.

Emerging Agentic Tools

Manus and Lemon

Manus has rolled out personal agents directly within Telegram, allowing users to execute multi-step tasks like research and document generation via chat. Meanwhile, Lemon launched a voice-to-task agent that lets users control their computer and delegate emails or research without typing. Check out Manus and Lemon.

Claude Code Desktop

Anthropic has released Claude Code on Desktop, a visual application that allows developers to run agentic coding tasks. It features parallel sessions, enabling multiple agents to work on different branches simultaneously. It is available for Pro, Max, and Team subscribers. Download it here.

Rapid-Fire Tool Updates

Tool

Description

Minimax M2.5 & GLM-5

New Chinese models showing strong coding and tool-calling benchmarks.

Gemini Deep Think 3

Google's new academic-focused model scoring 84.6% on ARC-AGI 2.

Recall.ai

Infrastructure for meeting bots used by Hubspot and Clickup.

Inngest Durable Endpoints

Middleware to make APIs unbreakable with one line of code.

ZVEC

Alibaba's open-source in-process vector database for rapid similarity search.

Wispr Flow

Voice-to-text tool that formats speech into final-draft writing.

HeyGen

Video generation platform for turning ideas into videos without filming.

#AI Tools#Qwen 3.5#OpenAI#Dreamer#Manus#Claude Code
Csaba Szirják
CTO & COO, AI Evangelist

Meet Csaba Szirják, the engineer behind testified.ai. With 20+ years as VP of Engineering, CTO, and WorldSkills Expert, Csaba audits AI software for enterprise integration, security, and ROI.