October has barely started and the open-weight side of the AI race is already looking more competitive than the closed-model side. Two launches this week — from Mistral AI and the Nvidia-backed startup Reflection AI — suggest that the gap between proprietary frontier labs and openly available models is narrowing fast, even as regulators start paying closer attention to what autonomous AI agents can do once they're out in the wild.

Mistral's 'Le Chonk' goes public

On October 6, French lab Mistral AI pushed out a public preview of Mistral Large 4, internally nicknamed "Le Chonk." Mistral AI announced the release of Mistral Large 4 as a public preview, describing it as a granular Mixture of Experts model with 1.05 trillion total parameters, 49 billion active per token, and a 1.6 billion parameter vision encoder. The release also integrates that vision encoder with support for function calling, structured outputs, document Q&A, batching, and Mistral's Agents and Conversations APIs.

What makes the launch notable isn't just the scale — it's the infrastructure story and the pricing. The model was trained from scratch on 3,800 Nvidia Grace Blackwell GPUs in Mistral's own European datacenters. The API went live at $1.36 per million input tokens and $4.18 per million output tokens, though the weights won't ship until the end of October, meaning self-hosting isn't yet possible. Interestingly, exact parameter counts have become a minor point of confusion between Mistral's own model card and secondary reporting, with some outlets citing a smaller 675B/41B configuration — a reminder that even in 2026, frontier model specs aren't always nailed down at launch.

Reflection AI's Beam aims at the open-weight throne

Just a day earlier, on October 5, Nvidia-backed startup Reflection AI launched its first open-weight model, Beam, claiming it rivals leading open-weight technologies from both the U.S. and China. The company said Beam performs on par with Chinese lab Z.ai's GLM-5.2 and is close to Alibaba's Qwen3.8-Max. Founded by two former Google DeepMind researchers, Reflection describes Beam as a "powerful workhorse model" built for coding, agentic tasks, and multi-step reasoning. Full weights and technical documentation are still pending later this month, but the positioning is clear: Western open-weight labs are explicitly framing their releases as answers to China's open-source momentum.

Regulators are watching the agents

Meanwhile, the policy backdrop for all this model velocity is getting tenser. The FTC opened an industry-wide investigation into Anthropic, OpenAI, and other AI labs to examine the risks their technology poses to consumers. It's the first U.S. enforcement effort built around rogue AI agents — autonomous systems that can take actions beyond what their operators intend — examined under existing unfair-or-deceptive-practices law rather than new AI-specific rules. The agency is drafting civil investigative demands, which function like subpoenas, and expects to send them in the coming weeks.

Taken together, the week paints a picture of an industry accelerating on two fronts at once: open-weight labs racing to match frontier capability without the closed-API gatekeeping, while regulators try to catch up with what happens when increasingly autonomous systems are let loose.

Sources

AI AgentsAI RegulationArtificial IntelligenceDaily AI UpdateMistral AIOpen Weight Models