Alibaba says its new open-source multimodal model, Qwen3.8-27B, passed 1M+ downloads within a few days of release, making it one of its fastest-growing models
THE SO WHAT
A 27B-parameter multimodal model crossing 1M+ downloads in days shows how quickly capable, open models can saturate developer stacks. If you’re paying premium rates for proprietary mid-size models, you now have to justify what you’re getting beyond what Qwen-class OSS delivers on local hardware.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIWarp’s new system is an out-of-the-box software factory for AI development
If Warp Factories actually makes “AI software factories” turnkey, the bottleneck shifts from infra plumbing to how opinionated your development workflow is. Teams that standardize on a factory pattern early will ship more experiments per dollar than those still hand-wiring agents and eval loops.
Applied AISnowflake’s Plan to Cut AI Costs
Snowflake’s focus on model routing and multi-model risk management is a clear signal that enterprise AI is entering its cost-optimization phase, not its “just ship something” phase. If your stack still assumes one default frontier model, you’re overpaying and under-hedged—start instrumenting usage and routing across smaller, cheaper models where quality allows.
Applied AIPerplexity’s free AI offer left it with millions more users in India
Perplexity’s Airtel promo shows that subsidized access can seed durable usage—India revenue up ~60% even as downloads fell once the offer ended. For AI assistants, distribution through telcos and OEMs may be a more powerful growth lever than marginal model improvements.
Applied AIThe economics of a voice agent: what a minute of conversation actually costs
Once you unpack ASR, LLM, TTS, orchestration, and telephony, “cost per minute” for voice agents is a composite of infra, latency targets, and failure handling—not a single rate card. Before you greenlight a call-center automation project, model worst-case concurrency and error rates or you’ll discover the real unit economics only after launch.