FlashProve operates in two phases: an offline phase that publishes a weight commitment taking 39 minutes for 16 million requests, and an online phase where GPUs produce proof for each request

Overview Debates Related claims Topics Sources

Nous Research released Arete-27B, an open-weight supervised fine-tune of Alibaba's Qwen3.8-27B designed to remove Chinese state framing from model responses

Deepgram and Daily committed to keeping the Pipecat framework open source and free under its existing BSD V2 license.

Pipecat is an open-source framework that orchestrates voice and video streams while linking speech-to-text, large language models, and text-to-speech components to minimize latency in AI applications.

Daily co-founder Kwindla Kramer chose Deepgram's acquisition offer over interest from other potential buyers because Deepgram CEO Scott Stephenson committed to keeping Pipecat open source and vendor-neutral.

Deepgram co-developed the Real-Time Context Standard open protocol with the Pipecat team to pass audio context metadata—such as whispers, tone, and model guidance—between voice agent components.

Deepgram and Daily are establishing an independent Pipecat Foundation governed by a steering committee that will include competing companies to manage the framework.

The Daily engineering team will join Deepgram as part of the acquisition, with Daily co-founder Kwindla Hultman Kramer serving as Deepgram's head of open source while Deepgram sells, supports, and develops Daily's existing product suite.

The Pipecat framework features over 200 integrations and almost 400 contributors, having been built by 18 engineers at Daily.

Baidu released ERNIE 5.5 Flash Base, a 280-billion-parameter text-only sparse Mixture-of-Experts foundation model that activates 14 billion parameters per token

Ricursive Intelligence released Ripple-65B Preview in three checkpoints: an instruction-tuned BF16 version, an NVFP4 version, and a BF16 base model

OpenAI stated that Pierre Computer Company's open-source software projects will remain open source following OpenAI's acquisition of the company

Tencent announced plans to release the model weights for Hy4 along with a detailed report describing Hy4's safety testing, as well as a faster, cost-efficient Flash version of Hy4

FlashProve achieves ~110 tokens per second on Llama 4 Maverick, whereas competing open-source zkML system VerInf yields ~0.02 tokens per second on the same model

FlashProve enables regulators to verify that approved models run on compliant hardware, helping enforce AI pacing rules and output provenance

FlashProve prevents AI providers from swapping in cheaper models to cut costs or using more expensive models to inflate benchmark performance

FlashProve generates mathematical proofs without exposing an AI model's internal weights, allowing commercial AI developers to publish weight fingerprints while protecting proprietary IP

FlashProve prover hardware requires 1 Nvidia H100 80GB GPU, 24 CPU cores, and 290 GB of host RAM, while capping request context at 1,024 tokens

FlashProve uses hash-based cryptographic proofs to compile interactive oracle proofs, providing post-quantum security guarantees similar to Starknet's L2 validity proofs

FlashProve verification requires no GPU and can be executed locally on a standard CPU

FlashProve allows third parties to cryptographically verify claims that specific tasks, such as vulnerability scans or business plans, were performed by specific AI models