# forck.live — 30 June – 6 July 2025 > AI lab announcements, news reports and podcasts. forck.live follows what companies announce about their own products. Press may corroborate an official story; it never leads and never appears as an item of its own. A missing date stays missing. A title is the publisher's words — where a company does not publish in English, an English rendering is listed separately as `Title in English` and named as this site's translation, never as theirs. This file is one issue of the weekly archive: every confirmed first-party announcement of ISO week 2025-W27, 30 June – 6 July 2025, newest first. The window never moves, so this URL always names the same seven days — but a post published inside them and detected later is added when it arrives, so a closed week can still gain entries. Every entry is one first-party announcement, and carries the same eight fields in this order before any others: Company · Announced · Category · Coverage · Announcement · Group · Source · Record `Source` is the company's own page the announcement was traced to and `Record` is this site's page for it. Both are safe to cite. `Coverage` is a number of outlets, or `not counted` — which means this record reached no count of its own, never that nobody wrote about it. `Category` is this tracker's own analysis, and is authoritative wherever it is stated. `not stated` means either that nothing has analysed the post yet or that the analysis placed it outside this taxonomy; neither is a verdict about the announcement. `Announcement` is whether the post is news of something the company did, rather than a how-to, a customer story, a changelog listing or a staffing note. `Group` is which of this record's groups the entry falls in, carried on the entry itself so it holds where the groups are headings and where they are not. One of: `models`, `covered`, `announcements`, `routine`. Anything after `Record` is held detail that only some announcements carry. - This week's page: https://forck.live/week/2025-W27 - The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md - Catalogue: https://forck.live/llms.txt ## Announcements ### TII — Announcing NeurIPS 2025 E2LM Competition: Early Training Evaluation of Language Models - Company: TII (tii.ae) - Announced: 2025-07-04T12:25:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/tiiuae/e2lm-competition - Record: https://forck.live/items/1669-announcing-neurips-2025-e2lm-competition-early-training-evaluation-of-language The NeurIPS 2025 E2LM Competition, focused on building benchmarks for early-stage evaluation of language models, particularly for scientific knowledge. Participants will use Small Language Models of 0.5B, 1B, and 3B parameters and submit solutions via Hugging Face. The competition includes multiple phases and cash prizes. ### Amp — Oracle - Company: Amp (ampcode.com) - Announced: 2025-07-04 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://ampcode.com/news/oracle - Record: https://forck.live/items/7830-oracle - Subject: Amp Amp added a new tool called oracle to its toolbox, which is a read-only subagent powered by OpenAI's o3. The oracle is designed to work alongside Amp's main agent, Sonnet 4, for reviewing, debugging, and analyzing code. ### Harvey — Integrating Deep Research into Harvey - Company: Harvey (harvey.ai) - Announced: 2025-07-03T17:10:32.900000+00:00 - Category: capability-change - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.harvey.ai/blog/integrating-deep-research-into-harvey - Record: https://forck.live/items/17702-integrating-deep-research-into-harvey - Subject: Harvey Harvey integrated OpenAI's Deep Research API into its platform within 12 hours of the API's release. The integration was enabled by Harvey's modular Workflow Engine and a culture of rapid prototyping with AI-assisted coding tools. The feature is undergoing internal testing and will be available through an early access period. ### Kyutai — Kyutai TTS 1.6B - Company: Kyutai (kyutai.org) - Announced: 2025-07-03 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://kyutai.org/blog/2025-07-03-kyutai-tts-1-6b - Record: https://forck.live/items/18353-kyutai-tts-1-6b - Subject: Moshi / Pocket TTS The nitty-gritty of Kyutai TTS 1.6B, our text-to-speech model. ### Kyutai — Kyutai TTS and Unmute now open-source - Company: Kyutai (kyutai.org) - Announced: 2025-07-03 - Category: open-weight-release - Coverage: not counted - Announcement: yes - Group: models - Source: https://kyutai.org/blog/2025-07-03-tts-unmute-open-source - Record: https://forck.live/items/18352-kyutai-tts-and-unmute-now-open-source - Subject: Moshi / Pocket TTS Kyutai has open-sourced its TTS 1.6B model and the Unmute speech enhancement system, releasing them with associated benchmarks and project details. ### Mistral AI — Announcing AI for Citizens - Company: Mistral AI (mistral.ai) - Announced: 2025-07-03 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://mistral.ai/news/ai-for-citizens - Record: https://forck.live/items/3320-announcing-ai-for-citizens - Subject: Mistral / Le Chat Mistral AI announced AI for Citizens. ### Genspark — WORLD FIRST: Genspark Introduces AI Docs That Create Any Document With One Prompt! - Company: Genspark (genspark.ai) - Announced: 2025-07-02 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.genspark.ai/blog/genspark-ai-docs - Record: https://forck.live/items/18277-world-first-genspark-introduces-ai-docs-that-create-any-document-with-one-prompt - Subject: Genspark ### Replit — Introducing Dynamic Intelligence for Replit Agent - Company: Replit (replit.com) - Announced: 2025-07-01T23:52:00+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://replit.com/blog/dynamic-intelligence - Record: https://forck.live/items/17964-introducing-dynamic-intelligence-for-replit-agent - Subject: Replit Agent Replit introduced three new capabilities for Replit Agent: extended thinking, high power model, and web search. These features can be toggled on or off per request. Extended thinking instructs the Agent to reason step-by-step, the high power model increases accuracy, and web search enables the Agent to fill knowledge gaps by searching the internet. ### OpenAI — No-code personal agents, powered by GPT-4.1 and Realtime API - Company: OpenAI (openai.com) - Announced: 2025-07-01T10:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/genspark - Record: https://forck.live/items/533-no-code-personal-agents-powered-by-gpt-4-1-and-realtime-api - Subject: GPT / ChatGPT / API OpenAI published a case study about Genspark, which built a $36M ARR AI product in 45 days using no-code agents powered by GPT-4.1 and the Realtime API. ### Factory AI — Agent Native Development - Company: Factory AI (factory.com) - Announced: 2025-07-01 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://factory.com/news/build-with-agents - Record: https://forck.live/items/17820-agent-native-development - Subject: Droid Factory AI published a guide introducing Agent-Native Development, a methodology for building software with autonomous agents that emphasizes precise specifications, small task scopes, and automatic verification. The approach involves a loop of exploring, planning, coding, and verifying, with humans retaining architectural decisions and oversight. The guide also recommends maintaining an AGENTS.md file in repositories to document build commands, codebase organization, and pull request requirements. ### LG AI Research — LG AI Research at the Forefront of Global AI Governance Discussions - Company: LG AI Research (lgresearch.ai) - Announced: 2025-07-01 - Category: safety-policy-update - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.lgresearch.ai/news/view?seq=572 - Record: https://forck.live/items/11003-lg-ai-research-at-the-forefront-of-global-ai-governance-discussions - Subject: EXAONE LG AI Research participated in the UNESCO Global Forum on the Ethics of AI 2025 in Bangkok as the only Korean company present, contributing to discussions on global AI ethics and responsible development. The company shared updates on its collaborative AI Ethics MOOC project with UNESCO, which aims to strengthen AI ethics capabilities in public and private sectors through educational programs for AI experts, researchers, and policymakers, partnering with institutions including Harvard University, New York University, and the UN University. ### LG AI Research — LG AI Research at the Forefront of Global AI Governance Discussions - Company: LG AI Research (lgresearch.ai) - Announced: 2025-07-01 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.lgresearch.ai/blog/view?seq=572 - Record: https://forck.live/items/4785-lg-ai-research-at-the-forefront-of-global-ai-governance-discussions - Subject: EXAONE LG AI Research participated in the UNESCO Global Forum on the Ethics of AI 2025 in Bangkok, discussed corporate responsibility in AI governance, and shared updates on its AI Ethics MOOC project with UNESCO. ### Replicate — The FLUX.1 Kontext hackathon - Company: Replicate (replicate.com) - Announced: 2025-07-01 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://replicate.com/blog/kontext-hackathon - Record: https://forck.live/items/2487-the-flux-1-kontext-hackathon - Subject: Platform Replicate announced a hackathon using FLUX.1 Kontext image editing model and Chatterbox TTS, with winners and project highlights; also noted that Black Forest Labs released FLUX.1 Kontext dev with open code and weights. ### Hugging Face — Training and Finetuning Sparse Embedding Models with Sentence Transformers - Company: Hugging Face (huggingface.co) - Announced: 2025-07-01 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/train-sparse-encoder - Record: https://forck.live/items/1670-training-and-finetuning-sparse-embedding-models-with-sentence-transformers - Subject: Platform To use the Sentence Transformers library to train and finetune sparse embedding models, including components like model, datasets, loss functions, training arguments, evaluators, and trainer class, with practical examples. ### Lovable — Introducing Agent Mode (Beta): enabling Lovable to think, plan, and take actions autonomously - Company: Lovable (lovable.dev) - Announced: 2025-06-30T17:00:00+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://lovable.dev/blog/agent-mode-beta - Record: https://forck.live/items/12139-introducing-agent-mode-beta-enabling-lovable-to-think-plan-and-take-actions - Subject: Lovable / AI app builder Lovable is rolling out Agent Mode (beta), which enables the AI to think, plan, and take actions autonomously. The upgrade reduces build error rates by 90% and introduces capabilities such as searching the codebase, reading files, inspecting logs, searching the web, and generating images. Agent Mode uses usage-based pricing, where requests can cost less than 1 credit or several credits depending on complexity. ### Sakana AI — Inference-Time Scaling and Collective Intelligence for Frontier AI - Company: Sakana AI (sakana.ai) - Announced: 2025-06-30T15:00:00+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://sakana.ai/ab-mcts - Record: https://forck.live/items/4711-inference-time-scaling-and-collective-intelligence-for-frontier-ai - Subject: Sakana models Sakana AI announces AB-MCTS, a new inference-time scaling algorithm that enables multiple frontier AI models to cooperate, achieving strong performance on the ARC-AGI-2 benchmark. The algorithm is open-sourced with code and a paper. ### OpenAI — AI in Australia—OpenAI’s Economic Blueprint - Company: OpenAI (openai.com) - Announced: 2025-06-30T07:00:00+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/global-affairs/openais-australia-economic-blueprint - Record: https://forck.live/items/534-ai-in-australia-openai-s-economic-blueprint - Subject: GPT / ChatGPT / API OpenAI, in partnership with Mandala Partners, released an AI Economic Blueprint for Australia, outlining a plan to boost productivity through AI. ### Microsoft AI — The Path to Medical Superintelligence | Microsoft AI - Company: Microsoft AI (microsoft.ai) - Announced: 2025-06-30 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://microsoft.ai/news/the-path-to-medical-superintelligence/ - Record: https://forck.live/items/10974-the-path-to-medical-superintelligence-microsoft-ai - Subject: MAI Microsoft AI introduced the Sequential Diagnosis Benchmark (SD Bench) based on 304 NEJM case records and the Microsoft AI Diagnostic Orchestrator (MAI-DxO). MAI-DxO paired with OpenAI's o3 achieved 85.5% diagnostic accuracy, compared to a mean accuracy of 20% among 21 experienced physicians. The orchestrator is model-agnostic and designed to emulate a virtual panel of clinicians. ### Suno — Suno Acquires WavTool - Company: Suno (suno.com) - Announced: 2025-06-30 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://suno.com/blog/suno-acquires-wavtool - Record: https://forck.live/items/10333-suno-acquires-wavtool - Subject: Music / Studio Suno announced its acquisition of WavTool, an AI-accelerated browser-based DAW, bringing its editing capabilities and core team into Suno. The WavTool team now holds leadership positions in product and engineering at Suno. ### Microsoft AI — The Path to Medical Superintelligence - Company: Microsoft AI (microsoft.ai) - Announced: 2025-06-30 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://microsoft.ai/news/the-path-to-medical-superintelligence - Record: https://forck.live/items/8439-the-path-to-medical-superintelligence - Subject: MAI Microsoft AI developed a sequential diagnosis benchmark (SD Bench) using 304 NEJM cases and an orchestrator (MAI-DxO) that combines multiple language models. The best-performing combination of MAI-DxO with OpenAI's o3 achieved 85.5% diagnostic accuracy, outperforming the 20% mean accuracy of experienced physicians. ### Baidu — Announcing the Open Source Release of the ERNIE 4.5 Model Family - Company: Baidu (baidu.com) - Announced: 2025-06-30 - Category: open-weight-release - Coverage: not counted - Announcement: yes - Group: models - Source: https://ernie.baidu.com/blog/posts/ernie4.5/ - Record: https://forck.live/items/4655-announcing-the-open-source-release-of-the-ernie-4-5-model-family - Subject: ERNIE Baidu announces the open-source release of the ERNIE 4.5 model family, a set of multimodal MoE models with up to 424B total parameters, available under Apache 2.0. The models achieve state-of-the-art performance on text and multimodal benchmarks.