This Week in AI: Open Weights and Open Wallets
Mistral, Reflection and Aleph Alpha all shipped open-weight models, while the funding rounds reached numbers that used to sound like science fiction.
Open weights kept winning quietly. Mistral opened a public preview of Large 4 on October 6, a 1-trillion-parameter multimodal model with 49 billion active parameters that scored 82 percent on a vulnerability-reproduction test where Claude Opus 5.5 and GPT-6 Astra scored near zero because they refused the task. The Apache-style weights are scheduled for late October. Reflection AI unveiled Beam on October 5, a 501-billion-parameter open-weight model aimed at coding and agentic work, and Aleph Alpha released Kolibri, a 78-billion-parameter European mixture of experts under Apache 2.0 with up to 1 million tokens of context.
AWS made Z.ai's GLM 5.3, a 744-billion-parameter open-weight coding model, generally available on Bedrock on October 5, giving regulated buyers a procurement path without a direct vendor contract. The 'run it yourself' option is no longer a hobbyist's compromise; for privacy-sensitive and repeated calls, a model on your own machine is now a real default.
The money was just as loud. ElevenLabs closed a $300 million employee tender that valued the company at $22 billion, roughly double its February number, with ElevenAgents now about 55 percent of revenue. Anthropic's IPO prospectus reportedly targets a valuation above $2 trillion, backed by roughly $518 billion in decade-long compute commitments and about $4.6 billion in 2025 revenue. OpenAI is said to be seeking more than $30 billion in new funding at a $1.4 trillion pre-money valuation.
TypeSafe, the startup behind Jev, a non-text decision model, raised $870 million at a $7.5 billion valuation led by Andreessen Horowitz, with Sequoia and DCVC, weeks after its September 15 launch, claiming a third of the Fortune 500 already use it for automation. The 'decide, don't generate' pattern is now properly funded.
The takeaway for builders is that open weights and deep pockets are no longer opposites. You can self-host a frontier-class model and still watch the hosted players raise nine-figure rounds, which means the real constraint is no longer access, it's judgment about where each belongs.