Filters
- Infron open-sources esp32-llm-service for on-device TinyStories inference on ESP32-S3
Infron open-sources esp32-llm-service, running a ~260k-parameter TinyStories model entirely on an ESP32-S3 with weights, tokenizer, runtime, and KV cache on-device, exposing a Chat Completions-compatible Wi-Fi API with SSE streaming at about 30 tokens/s in its hardware record.
- Infron open-sources a testing suite for NVIDIA DGX Spark GB10
Infron says it has open-sourced its NVIDIA DGX Spark GB10 Testing Suite, arguing that higher throughput does not mean faster responses for every user and that peak specs differ from real-world latency.
- Infron launches a Trust Center for reviewing model-provider policies and security controls
Infron opened a Trust Center where customers can review Model × Provider policies, security controls and trust documentation for multi-provider AI routing. Infron says trust depends on the route, not only the model.
- Infron presents at a VIP dinner during the SF AI Summit: The Infrastructure Era
Infron says Vanessa Lu represented the company at a VIP dinner during the SF AI Summit: The Infrastructure Era, where it presented and exchanged ideas with leaders across AI, and thanked B.E.L.L.E for bringing the community together.
- Infron adds seven cybersecurity and abliterated models, including GLM-5.3 Flash Cybersecurity, behind one API
Infron says seven models for authorized security research and defensive use are now on its OpenAI-compatible API: DeepSeek V4 Flash Vision Exp Cybersecurity, Gemma 4 26B A4B IT Cybersecurity, NVIDIA Nemotron 3.5 Content Safety, Qwen3.8 27B Cybersecurity, GLM-5.3 Flash Cybersecurity, GLM-5.3 Abliterated and GLM-5.3 Flash Abliterated. It says teams can compare them on the same workload without separate provider integrations.
- Infron offers nine free models including DeepSeek V4, Qwen3.8 and Ling 3.1
Infron says nine models are free on its platform at $0 per million input and output tokens, including DeepSeek V4, Qwen3.8, Atria Dawn, Ling 3.1 and Apodex 1.1 Mini, so users can build coding, reasoning and multimodal prototypes without spending credits.
- Infron launches Batch API for async bulk inference at up to 80% lower cost
Infron says its Batch API is live for serverless async bulk jobs aimed at evals, synthetic data and labeling, claiming up to 80% lower cost than its standard workflow.
- Infron adds seven low-refusal models for authorized red-teaming and security research
Infron has made seven low-refusal models available, aimed at authorized red-teaming, vulnerability research and complex multi-step execution. They include Mistral 24B Venice Edition, Gemma 4 31B Abliterated V2, Qwen3.5 9B Abliterated, Qwen3.8 27B Abliterated, SuperGemma 4 26B Uncensored, GLM 4.7 Flash Heretic and Gemma 4 26B A4B Uncensored. Infron says they are for authorized and responsible use only.
- Infron lists upcoming AI infrastructure and agent events in San Francisco and Menlo Park
Infron announced upcoming events for AI and infrastructure builders: SF AI Summit, The Infrastructure Era during SF Tech Week; Agents in Production on October 20 in Menlo Park; and Open Infra Night for Agent Builders on October 22 in San Francisco.
- Cua Driver adds an optional Perception visual extension
Cua added an optional Perception visual extension to Cua Driver, which previously relied primarily on web structure and system accessibility trees when driving apps.
- Infron introduces Priority tier for GPT-5.6 Luna
Infron introduced a Priority tier for GPT-5.6 Luna aimed at latency-sensitive chat, classification and lightweight agents, with less queueing and steadier capacity, per its official X account.