Lookonchain APP

App Store

Prime Intellect Acquires Moondream: High-Speed Inference Engine Built by Fewer Than 10 People

50 minutes ago

Beating AI Express News: AI infrastructure firm Prime Intellect, valued at $1 billion, announced the acquisition of open-source vision model company Moondream. The latter has a team of fewer than 10 people, previously raised $4.5 million in funding, with the acquisition amount undisclosed. Both founders will join Prime Intellect to lead inference acceleration and multimodal model research. Moondream’s models can identify warehouse bins, locate products on shelves, and analyze live video. The latest Moondream 3.1 has 9 billion total parameters, with only 2 billion activated per run; the entire model family sees over 5 million downloads monthly. However, the more critical asset in this acquisition is the Photon inference engine, which consolidates large volumes of scattered GPU compute operations to reduce hardware latency. Prime Intellect stated that on a single H100 GPU, Photon can run over 60 Moondream inferences per second, sufficient to support live video analysis. Prime Intellect previously focused on providing infrastructure for AI agent training, reinforcement learning, and other use cases, and launched its Prime Inference platform earlier this October. Following the acquisition, Moondream 3.1 has been integrated into the platform, and Photon has begun providing inference acceleration for Clef and Clef Flash. The company plans to further apply this technology to video, voice, and computer operation AI agents, enabling enterprises to complete multimodal model training, optimization, and deployment on a single platform. The Moondream brand will be retained.

Relevant content

US optical module and storage stocks fall in pre-market trading.

According to Bybit market data, US-listed optical module and storage stocks were trading lower in pre-market hours, with the following movements: COHR down 0.27%, LITE down 0.66%, AAOI up 0.22%, NOK up 1.52%, MRVL down 0.31%, STX (Seagate) down 1.47%, WDC (Western Digital) down 1.24%, SNDK (SanDisk) down 2.16%, and MU (Micron) down 0.63%.

18 minutes ago

DeFi yield protocol infiniFi unveils INFI token economic model, with 45% allocated to the ecosystem.

According to official announcements, DeFi yield protocol infiniFi has unveiled the token economics model for its native INFI token, with the Token Generation Event (TGE) on the horizon. INFI’s total token allocation breaks down as follows: 45% for the ecosystem, 20% for the team, 20% for investors, and 15% for the foundation. infiniFi stated that its upcoming community airdrop will be allocated from the ecosystem pool. More use cases for INFI will be announced ahead of the TGE, while additional details on the token economics model, allocation mechanisms, and the airdrop are set to be disclosed in the near future.

18 minutes ago

US crypto-related stocks rose slightly in pre-market trading.

According to Bybit market data, as Bitcoin rebounded to briefly top $84,000, US-listed crypto-related stocks posted modest pre-market gains, with SBET rising 0.83%, BMNR up 0.64%, GEMI up 0.49%, CRCL up 0.64%, MSTR up 1.49%, COIN up 0.57%, and ASST up 0.94%.

18 minutes ago

Liquid AI upgrades its 24B-parameter local model: only 2B parameters are activated, and it outperforms the larger Qwen model in agent tests.

Beating AI Express: Leading edge-side AI R&D firm Liquid AI has launched LFM2.5-24B-A2B, a model focused on enhancing local agent inference and tool calling capabilities. The previous LFM2-24B-A2B generation prioritized operational efficiency; the new version retains a total of 24 billion parameters with only around 2 billion activated per run, while its pre-training data has grown from 17 trillion to 33 trillion tokens, and its context window expanded from 32K to 131K. The new model performs inference before generating responses, and was fine-tuned via reinforcement learning directly on agent frameworks including OpenClaw, Hermes Agent, Pi, and OpenCode, training on continuous tool calls, handling return results, and completing multi-step tasks. In an official demo, a MacBook—no cloud model API required—enabled Hermes Agent to search for trending Hugging Face projects, compile briefings, and act on follow-up queries. Official tests show the new model generates 184 tokens per second on Apple M5 Max, up to 1.7 times faster than comparable MoE models. In agent benchmarks, it scored 48.09 on LiveClaw, higher than Qwen3.6-35B-A3B’s 38.19; its AutomationBench score of 27.69 is over three times that of Qwen. However, it only achieved 51% on the SWE-bench Verified programming test, lower than Qwen’s 66.2%.

18 minutes ago

Binance: Users with at least 245 points can claim the Alpha airdrop at 18:00 today.

According to official announcements, Binance Wallet will open the claim period for a new round of Alpha airdrops at 18:00 Beijing time today. Users holding at least 245 Binance Alpha points are eligible to claim the token airdrop. The airdrop operates on a first-come, first-served basis and will remain open until the airdrop pool is exhausted or the event concludes.

18 minutes ago

TRUMP team's allocated address transfers 6.136 million TRUMP tokens to a new address.

According to on-chain data analyst @ai_9684xtpa’s monitoring, TRUMP’s team allocation address transferred $11.78 million worth of tokens (6.136 million TRUMP tokens) to a new address after a 12-day interval, likely to be deposited into a trading platform. It is noted that 12 days ago, this address had already transferred $8.87 million worth of TRUMP to the address 52RDq…rkaD1; these tokens were later deposited into OKX via two addresses, using a similar operational approach.

18 minutes ago

Popular tokens

BitcoinEthereumHyperliquidSolanaTRONBNBTetherAaveXRPPepeFartcoinOndoJupiterUniswapBonkPendleEthenaArbitrumAvalancheLidoChainlinkPolygonDogecoinCardano