In partnership with
China's DeepSeek spent the week making AI easier to customize and cheaper to run, which puts fresh pressure on the US labs spending fortunes to stay ahead. We checked the releases and the filing behind the headlines, and labeled the figures that come from the companies themselves.
100+ Claude Code hacks to ship code 10X faster
Top engineers at Anthropic say AI now writes 100% of their code.
Are you using AI to write yours?
These 100+ Claude Code hacks show you exactly how. Sign up for The Code and get:
100+ Claude Code hacks — free
The Code newsletter — learn the latest AI tools and skills to code faster in 5 mins a day
TODAY IN AI
3 things that happened while you were busy
1. DeepSeek lets you build custom AI tools just by describing them.
DeepSeek showed off Creator Mode, a feature of its open-source Harness platform that lets you build your own plugins by making plain-language requests, no coding required. The demos included a paper reader, a poster designer and a stats dashboard, each refined over a few rounds of conversation. Read TechNode's report. It pushes custom AI toward people who cannot code, which is where most of the market actually is.
2. DeepSeek open-sourced a coding agent aimed at Claude.
DeepSeek also released a free, MIT-licensed prototype of a coding "super-agent" that takes an authorized task and returns the work, often as a GitHub pull request, framed by Bloomberg as a shot at Anthropic's Claude Code. The public version is a stripped-down prototype, not DeepSeek's full internal system. See the details. Claude Code reportedly runs at over $2.5 billion a year, so a free rival is a direct challenge.
3. Anthropic's updated filing shows huge revenue and a bigger bet.
Anthropic's amended IPO filing reported about $17 billion in third-quarter revenue, up from $11.6 billion the quarter before, but a return to an operating loss, alongside roughly $895 billion in compute purchase commitments and leases. Those figures are from the filing as reported by Fortune and TBPN. See the breakdown. The gap between soaring revenue and that spending is why every quarter now gets scrutinized.
Replace your traditional insurance broker.
WithCoverage replaces your company’s traditional insurance broker with an AI-enabled team of insurance experts and attorneys. A few things that make us better:
40+ carriers compete for your business instead of 2 or 3.
We operate on a Flat Fee instead of broker commissions
You get a platform to manage policies, claims, COIs, etc.
1000+ businesses (Gopuff, Polymarket, Opendoor) replaced their traditional broker with us.
Let us review all your policies for free to show you how we can save you money and better protect your business.
Your broker will not be notified. This is simply our way to earn your trust.
TODAY'S TUTORIAL
Build your own custom AI assistant, no code needed
You do not need DeepSeek's Creator Mode to make a custom AI helper. ChatGPT, Claude and Gemini all let you build one for a repeat task in minutes. Here is how to make one that handles a job you do every week. See the trend.
1. Open the builder. In ChatGPT use "Create a GPT," in Claude make a Project, in Gemini make a Gem. Each one is a saved assistant with its own instructions.
2. Name the job. Pick one clear task, like writing product descriptions or replying to common customer questions, instead of a do-everything bot.
3. Write the instructions. Tell it who it helps, the steps to follow, the tone to use and what never to do. The prompt in the Prompt Station below writes this part for you.
4. Feed it examples. Upload a few of your best past examples or reference files so it matches your style and facts rather than guessing.
5. Test and reuse. Run it on a real task, fix the instructions where it drifts, then use the same assistant every time instead of rewriting a prompt.
Good to know: Custom GPTs need a paid ChatGPT plan to build, though Claude Projects and Gemini Gems are available on their free tiers with limits.
FROM THE FRONTIER
Cheap, open AI from China is setting the price for everyone
The pressure. DeepSeek keeps shipping open, low-cost models and now agent tools, giving away capability that US labs sell. When a free option is close enough, it drags the whole market's prices down. See the agent.
The spend. Meanwhile Anthropic's filing shows around $895 billion in compute commitments against a quarter that slipped back into a loss. Staying at the frontier is getting more expensive, not less. See the numbers.
The opening. For a small operator, this is good news. Open weights you can run on your own hardware and APIs priced in cents mean you can build real tools without a big lab's budget. See the cost gap.
The reality check. Benchmark wins are mostly the makers' own claims, open models still need real hardware to run, and using Chinese models raises data and policy questions for some businesses. Test before you trust. See the caveats.
The reframe. You are not picking a side in a US-China race. You are the customer a price war is fought over. Keep your work portable across models, and let the giants burn cash to win you.
MONEY FORWARD
Three ways to turn this week's AI news into income
1. Build custom AI assistants for clients. Most small businesses want a bot for one job but do not know how to set one up. Use the tutorial above to build Custom GPTs or Claude Projects for them and charge per build plus a small retainer. See the demand.
2. Cut a client's AI bill. With open and low-cost models this good, many businesses overpay for premium APIs they do not need. Offer to move suitable workloads to cheaper models and split the savings. See the pricing.
3. Set up a coding or ops agent for a small team. Free agent tools now handle routine pull requests and tasks. Configure one for a small dev shop or store, with the human approvals in the right places, as a paid setup. See the tools.
GET THE EARNINGS GUIDE
Want the step-by-step version of ideas like these? Get The Aiwire earnings guide at theaiwire.co. These are ideas, not income promises, and results depend on your skills, market and effort.
IN THE KNOW
What people are actually watching and sharing
China's cheap frontier model. DeepSeek's V4-Flash scores near proprietary models on independent benchmarks at a fraction of the cost per task, with free weights under an MIT license and an API around $0.14 per million input tokens. See the benchmarks.
A US open-weight answer. Meta released Muse Spark Open, which it calls the most capable US open-weight model so far for coding and agent tasks, light enough to run on a pair of Mac Studios, per a roundup of the week's releases. See the release.
Claude gets a background agent. Anthropic opened a research preview of Claude Duet, an assistant that watches your connected apps and asks for approval before important actions, on a waitlist for US Pro and Max subscribers. See the preview.
Many agents at once. OpenAI's GPT-6 Pro Parallel Agents splits a single task across several agents working at the same time inside ChatGPT, aimed at bigger, multi-step jobs. See the feature.
Microsoft's own voice stack. Microsoft is building out in-house voice and transcription models, MAI-Voice-2.1 and a streaming transcribe model, reducing its reliance on partner models for speech. See the list.
TRENDING TOOLS
5 tools worth a look this week
1. DeepSeek V4-Flash. A free, open-weight model that is close to the top models on many tasks, with an API priced in cents. Good for cheap automations if you are comfortable with a Chinese model. See the model.
2. Claude Duet. Anthropic's background agent that monitors your connected apps and acts with your approval. A research preview on a waitlist for US Pro and Max users, on iOS, Android and the web. See the preview.
3. Meta Muse Spark Open. A free open-weight model for coding and agent work that you can download and run on your own hardware, including two Mac Studios. No API bill once it is set up. See the details.
4. Mistral Large 4. An open-weight, roughly trillion-parameter model that is strong at coding and free to download if you have the hardware. Carried over from last week as a standout. Read the details.
5. ElevenLabs SFX v3. Generates sound effects from a text prompt and blends several into one clip, handy for short video and podcasts, with a free tier. Still one of the most useful creator tools around. See the tool.
PROMPT STATION
Design your own custom AI assistant
This prompt writes the instructions for the custom assistant in today's tutorial. Run it in Claude or ChatGPT, then paste the result into a Custom GPT, a Claude Project or a Gemini Gem. It turns a job description into a working setup.
COPY AND PASTE THIS PROMPT
I want to build a custom AI assistant for this one job: [DESCRIBE THE TASK, e.g. writing product descriptions for my Shopify store]. The people it helps are [WHO], and it should sound [TONE]. Write the full setup for it: 1) a one-line description, 2) the step-by-step instructions it should follow every time, 3) the rules of what it must never do, 4) three example questions it should handle well, and 5) a list of what I should upload as reference so it matches my style and facts. Ask me for anything you need before writing it.
Tip: Swap [DESCRIBE THE TASK] for one specific job, not your whole business. For a sharper assistant, paste one example of a great result you have made before and tell it to treat that as the standard to match.