Open-source Adobe clones, on-device AI, and a new speech recognition model
How open-source tools challenge Adobe's price, why small AI models now run without the cloud, and a new speech-to-text model for transcription.
Summary
- Open-source Adobe clones — a developer released open-source clones of Adobe’s tools; a good reason to review what you actually pay for before your next renewal.
- Small AI models on-device — Liquid AI d1 now accepts images and responds in tens of milliseconds without the cloud, good for fast sorting tasks.
- Falcon ASR — a new speech recognition model — TII’s speech recognition model that turns speech into text; mainly Arabic (Emirati) plus four other languages, with only a demo available so far.
The underlying feel: this week is dominated by one theme — expensive, subscription-locked tools getting free, open-source competitors, in both design and speech recognition.
1. Open-source Adobe clones — a reason to review your software subscription
What changed. A developer released open-source clones of Adobe’s tools, openly targeting the company, claiming that “software is over”.
What it means for your business. If your team pays for an Adobe subscription but uses only part of the features — image editing, PDF work, some design tasks — then it’s worth keeping an eye on these open-source clones. How well they actually replace Adobe’s tools is not something the source states, so test them yourself before relying on them.
What you can do this week.
- Write down exactly how many Adobe tools your team really uses per month — often you pay for the full suite because of a third of the features.
- For simple image and design work, try free tools like GIMP or Photopea (runs in the browser, €0) — check if they cover your most common tasks.
- Before your next renewal, compare: the full suite’s cost versus how much work you can actually do in free tools.
What to watch.
- More open-source alternatives to expensive creative tools will appear in the coming months.
- Subscription providers will likely push cheaper plans harder to retain customers.
2. Small AI models that run on the device itself
What changed. Liquid AI released open d1 decision models for the edge that accept images and run directly on the device — without a constant cloud connection.
What it means for your business. If you process many applications, images or documents and each call to a large cloud model costs money and takes time, running a small model on the device itself means lower costs and faster responses. For example, if you sort incoming requests by category or check whether a document has the right information — a small model can now do this locally.
What you can do this week.
- If you have a technical staff member, ask them to try the Liquid AI d1 model on a simple sorting task — the model is open and free.
- For simpler local AI tasks without a subscription, try LM Studio or Ollama (€0) — they let you run small models on your own computer with a few minutes of setup.
- Identify which repetitive tasks (sorting, categorizing) currently go through an expensive cloud model and whether they can move to a cheaper local one.
What to watch.
- Over the coming months more everyday AI tasks will move from the cloud to the device, cutting monthly costs.
- Ready-made tools will appear that make these small models usable without a technical staff member.
3. Falcon ASR — a new speech-to-text model
What changed. TII (Abu Dhabi) introduced Falcon ASR — a speech recognition model that turns speech into text, focused on Arabic (Emirati dialect) and also supporting English, French, Spanish and Portuguese. For now only a demo is live, with API access and apps planned.
What it means for your business. If your team writes meeting notes, transcribes client calls or creates records from voice recordings — that’s time-consuming manual work. Open-source speech recognition means this can be automated without a subscription and without recordings going to a third-party service. For example, if you run a legal or consulting practice where call content must be documented, transcription is no longer hours of manual work.
What you can do this week.
- If you regularly transcribe meetings, try OpenAI Whisper (open, €0) — record one meeting and check quality; you can only try Falcon ASR as a demo on its page for now.
- For a quick free option without setup, try Google Docs voice typing or MacWhisper (partly free) — good for short notes.
- Build a simple routine: record → transcribe → AI summary in ChatGPT (free plan) — the whole process goes from an hour to a few minutes.
What to watch.
- Speech recognition will keep improving for smaller languages — worth testing quality with each new model.
- Meeting tools (e.g. Teams, Zoom) will include transcription and summaries as standard at no extra cost.
Today’s picture
All three stories share one structural change: expensive, subscription-locked tools are getting free, open-source competitors. Design software, cloud AI calls and paid transcription services — each got a new, more open competitor this week. For a business owner this means one concrete action: review which tools you currently pay a fixed subscription for, and check whether a free alternative covers your most common work.
| Event | Consequence |
|---|---|
| Open-source Adobe clones | Creative software gets open-source clones — a reason to review subscriptions |
| Liquid AI d1 on-device | Repetitive AI tasks can move from an expensive cloud to a cheap local model |
| Falcon ASR | A new speech recognition model; only a demo is available for now |
3 questions for the reader:
- Which software subscription haven’t you reviewed this year, even though you use only part of the features?
- How many hours per week does your team spend manually transcribing meetings or calls?
- Which repetitive AI tasks in your work currently go through an expensive cloud model?
Also briefly
- Nvidia physical AI — Nvidia is betting on physical AI, aiming for safer robotaxis and humanoid robots, reports Ars Technica (3 min read). Worth following if your industry touches logistics or transport.
- Microsoft’s new hardware — Microsoft’s event debuted new AI-friendly hardware and Windows changes, writes Ars Technica. Worth knowing before your next computer refresh.
- Nemotron at gold level — NVIDIA’s Nemotron model family reached gold-level results in two math and coding olympiads (IOI and IMO), described by HuggingFace (5 min read). A signal of how far open models have come on complex tasks.
- AI music fraud — a man was jailed for using 10,000 bots and AI songs to artificially inflate stream counts, reports Ars Technica (1 min read). The first criminal case of AI-assisted music streaming fraud, investigated by the FBI.
- AI security incidents — a study collects lessons from OpenAI, Anthropic and Google agent security incidents, shifting from reactive to proactive, explains ArXiv (1 min read). Useful if you plan to deploy AI agents with data access.
Sources
- Ars Technica — “Software is over”: open source Adobe clones
- HuggingFace — Multimodal open d1 decision models for the edge (3 min read)
- HuggingFace — Introducing Falcon ASR (4 min read)
- Ars Technica — Nvidia’s big bet on physical AI (3 min read)
- Ars Technica — Microsoft event debuts new AI-friendly hardware
- HuggingFace — Nemotron IOI and IMO (5 min read)
- Ars Technica — Fraudster jailed for AI songs streaming fraud (1 min read)
- ArXiv — From Reactive Containment to Proactive Assurance (1 min read)
AI news for business leaders
Twice a week: what AI news means for your business. No hype, with concrete steps.