Iran's Foreign Minister: Negotiations are still ongoing, and a response to the U.S. proposal will be provided in the coming days.

US initial jobless claims for the week ending October 3 totaled 197,000, compared with market expectations of 200,000, while the prior week’s figure was revised from 197,000 to 199,000.

GlobalFoundries (GFS) and TSMC have signed a $2 billion silicon interposer agreement with an initial 5-year term. According to market data from BIT (bit.com), GlobalFoundries (GFS)’s pre-market gain on US stocks has widened to over 5%.

According to market data from BIT (bit.com), Google's US stock flipped from negative to positive in pre-market trading, now rising 1%. At the Gemini at Work 2026 conference, Google Cloud CEO Thomas Kurian unveiled Gemini Agent, positioned as a single universal work agent. Users only need to state their goal rather than specific instructions to accomplish knowledge work, content creation, image and media generation, as well as code writing and execution.

Beating AI Express: US enterprise payment platform Ramp has released its latest software procurement ranking, based on real transactions from over 70,000 businesses on its platform. TypeSafe AI topped the "Explosive Growth" list with its Jev model. In the market share growth ranking, it also placed second, trailing only Anthropic, while OpenAI ranked fourth. Jev, launched less than a month ago, has seen an approximate 1 percentage point rise in enterprise adoption. It is an AI model specifically designed to help software make judgments, not generating chat responses, but handling classification, selection, and scoring tasks. It charges $0.042 per million input tokens, with outputs provided free of charge. Low-cost models or model access platforms including DeepSeek, Featherless AI, and OpenRouter also feature on the list. However, demand for high-priced AI tools among enterprises is also growing. Video generation platforms Runway and Higgsfield have both entered the top ten in market share growth, and are currently primarily used in advertising and social media marketing. Ara Kharazian, chief economist at Ramp, stated that low-cost models may lower enterprises' AI call costs, but rising demand for image and video generation could push up overall AI spending.

According to HTX market data, STRK (Starknet) rallied 18.1% in 24 hours, decisively breaking above the $0.06 mark. On the news front, StarkWare CEO Eli Ben-Sasson proposed at Token2049 that blockchains need post-quantum readiness and cryptographic agility to counter threats from quantum computing and AI. Ben-Sasson noted Starknet is considering multiple options, including converting to a Layer 1 (L1) chain to independently advance security migration efforts. Earlier developments include: On October 2, the strkBTC incentive program covered bridging fees for the first 100 bridged BTC and distributed weekly faucet rewards; On October 5, the v0.14.4 mainnet launched, expanding SNIP-36 single-proof capacity to approximately 1.1 billion L2 gas.

Beating AI News: Perplexity has open-sourced two multimodal embedding models, pplx-embed-v2-late, with sizes of 0.6B and 9B parameters respectively. The two models are vector-compatible, allowing developers to use the larger 9B model for building knowledge bases while leveraging the smaller 0.6B model for daily searches—eliminating the need to run the large model for every query. The new models support text, image, and PDF page retrieval. Traditional embedding models typically compress content into a single vector, which often leads to detail loss. In contrast, the new models retain a 128-dimensional vector for each token, enabling search queries to match relevant segments within documents directly. When processing PDFs, PPTs, and scanned materials, the models can convert page images straight to vectors, bypassing OCR text extraction while preserving charts, tables, and layout information. Both models are trained from the same 18B teacher model and share a common vector space. In official ViDoRe v3 image retrieval tests, the 0.6B model scored 62.3% when used for both database building and queries; switching to the 9B model for database building and the 0.6B model for queries boosted the score to 63.5%; the full 9B model achieved 65.2%. The model weights are now available on Hugging Face under the MIT license.