MiniMax says M3 is available for coding and agent work, but the public repo still says the model is not released. Builders should treat this as staged access, not a finished open release.
Jun 9, 2026·3 min read
Anthropic’s new model is pitched as a coding upgrade, but the real product is control over how much agent work costs before it runs away.
May 31, 2026·4 min read
DeepSeek’s V4 preview is less interesting as a benchmark claim than as a pricing and product move: long context is becoming table stakes.
May 29, 2026·3 min read
Mistral Medium 3.5 and GPT-5.5 fold everything into one endpoint. When does a single model outperform a router with three specialists? It depends on latency, cost, and task distribution.
May 7, 2026·7 min read
Mistral folded chat, reasoning, and code into one 128B dense model. The question is whether the simplicity actually wins in production, and when it does not.
May 3, 2026·8 min read
Mistral AI launched Medium 3.5 on April 29, 2026, a 128-billion-parameter dense model that merges instruction-following, reasoning, coding, and vision into a single set of weights. It replaces Mistral
May 3, 2026·8 min read
OpenAI has stopped reporting SWE-bench Verified scores, arguing the benchmark no longer measures real coding progress so much as data leakage and broken tests.
Apr 26, 2026·6 min read
OpenAI released GPT-Rosalind, a purpose-built model for life sciences and drug discovery. The signal is clear: the next era of AI competition belongs to domain specialists.
Apr 26, 2026·6 min read
Claude Mythos Preview outperforms Opus 4.7 on every published benchmark. Instead of shipping it, Anthropic put it behind a Cyber Verification Program. Nobody has fully explained why.
Apr 24, 2026·5 min read
GPT-Rosalind is OpenAI's first vertical-specific model, built for life sciences. It signals a shift from general-purpose AI to specialized systems.
Apr 24, 2026·6 min read
GPT-5.5 is OpenAI's pitch for real work: better coding, stronger tool use, and much longer context, with higher prices and real reliability tradeoffs.
Apr 24, 2026·5 min read
GPT-5.4-Cyber opened access to thousands. Opus 4.7 locked Mythos behind a vetting program. One week, two entirely different answers to the same question.
Apr 23, 2026·5 min read
Moonshot AI released Kimi K2.6 with 300 sub-agents coordinating simultaneously. The agent swarm race is no longer a US story. China's open-source release changes the competitive dynamics.
Apr 23, 2026·5 min read
DeepSeek V4 runs on Huawei chips, not Nvidia. It's the first trillion-parameter frontier model built entirely outside US export controls. The implications go far beyond technology.
Apr 21, 2026·5 min read
An Artificial Dispatch investigation into the realities of artificial intelligence adoption, capability, and consequence.
Apr 4, 2026·5 min read
The Briefing