DeepSeek Puts Its App Inside the API, Antigravity Splits Planning From Doing, and CALL-E Hands Agents a Phone Line
DeepSeek opened the search, file, voice and safety tools behind its consumer app to API developers, Google paired a planning model with Gemini 4 Hydrogen in a new Antigravity Orchestrator mode, and AI Rudder launched CALL-E so assistants like ChatGPT can place real phone calls after the user approves a call plan. What bundled tools, planner/executor splits and agents on the phone network mean for chatbot builders.
This week's launches weren't about one bigger model. They were about wiring: which tools ship with the model, which model does which part of a job, and how far an agent can reach outside the chat window. DeepSeek bundled its app's tools into the API, Google split one coding session across two models, and a Singapore startup gave chat assistants the ability to dial real phone numbers. Most details come from company announcements and launch-week coverage, so treat specifics as provisional.
1. DeepSeek Ships the App as an API
On November 13, DeepSeek made the features behind its consumer app callable from its API: web search, link reading, file parsing, voice and a safety check. Coverage says they work across the Messages, Responses and Chat Completions formats, that the search tool is named deepseek_web_search, and that only the Flash model can use them for now. With thinking turned on, the model can search and read links over several reasoning rounds. Without it, search behaves differently. Speech-to-text, text-to-speech and OCR are also offered as standalone APIs. It builds on the Search and Fetch API DeepSeek opened in October.
Reported prices are low. Search is about $0.75 per 1,000 successful requests and OCR about $0.225 per 1,000 pages. Speech-to-text and text-to-speech are both at half price for a limited time, with no end date published. Every tool fee is added on top of normal token charges. These figures come from secondary write-ups that cite DeepSeek's pricing page, and they don't all agree, so check the page itself before you budget.
The trend is that the big providers now sell the whole assistant (retrieval, documents, voice, moderation) and not just the model. That saves glue code, but your bill now has several more lines, and a built-in search tool decides what counts as a source for you. Log each tool call with its cost the same way you log tokens:
type ToolCharge = { tool: 'search' | 'ocr' | 'asr' | 'tts'; units: number; usdPerUnit: number };
function turnCost(tokensUsd: number, charges: ToolCharge[]) {
return tokensUsd + charges.reduce((sum, c) => sum + c.units * c.usdPerUnit, 0);
}
// Alert when tool fees, not tokens, start to dominate a conversation's cost.
2. Orchestrator: One Model Plans, a Faster One Does the Work
Alongside Gemini 4 Hydrogen, Google added an Orchestrator option to the model picker in its Antigravity coding environment. A frontier model (Carbon, or Argon in a second variant) writes the plan, hands code edits and tests to Hydrogen, and steps back in to fix the plan when execution runs into trouble. Google says it got close to Carbon's own success rate on Terminal-Bench while using about 24% less quota and finishing about 19% faster. Those are vendor numbers, and the 24% is quota inside Antigravity, not money off a bill. Sources disagree on which plans get which variant, but every report puts the Carbon version on Ultra. One user reported that picking Orchestrator also switches approvals to auto-accept by default.
The planner/executor split is now a product feature, not just a pattern from research papers. If you route chatbot traffic, the lesson carries over: send the expensive model the decisions and let a cheap one carry out well-defined steps. Make the handoff explicit, too. That means a written plan the executor follows, plus a clear trigger for escalating back to the planner. And check what a mode changes besides the model. A setting that also loosens approvals is a security change.
3. CALL-E Takes Agents Onto the Phone Network
AI Rudder, a Singapore contact-center AI company, launched CALL-E on November 13 and 14. It lets assistants like ChatGPT, Meta's Muse and OpenClaw place real phone calls for their users: booking a table, getting a quote, moving an appointment or checking stock. The listing describes calls to US and Canadian numbers. The agent asks for any missing details, shows the user a call plan, and dials only after the user approves it. The company says a pre-call check blocks greetings that pose as emergency services or law enforcement. After the call, the user gets a summary, a confidence assessment, the transcript lines that support it, and a record linking the plan to the result. All of these safeguards are the company's own description.
For builders, the useful part is the receipt, not the dialing. When an agent acts somewhere the user can't watch, it should leave behind evidence the user can check. That means what was planned, what actually happened, and how sure the agent is.
Conclusion
Agents are getting more capability through tool bundles, split models and new channels. Each one adds a place where cost, permissions or accountability can slip. Meter every tool, make the hand-offs between models explicit, and require a plan before the agent acts and a receipt after.
Suggested visuals: a stacked bar showing one research conversation's cost split between tokens and DeepSeek tool fees; a sequence diagram of an Orchestrator session from Carbon's plan through Hydrogen's edits back to re-planning; and a CALL-E flow from call plan to user approval, impersonation check, call and transcript-backed summary.
— Maya
Frequently asked questions
Which DeepSeek model can use the new built-in tools?
Coverage of the November 13, 2026 release says only the Flash model can use web search, link reading, file parsing and the safety check for now. Speech-to-text, text-to-speech and OCR are also sold as standalone APIs, and tool fees are billed on top of token charges.
What is Orchestrator mode in Google Antigravity?
It is a model-picker option that pairs a planning model (Gemini 4 Carbon or Argon) with Gemini 4 Hydrogen for execution in one session. Google reports near-Carbon results on Terminal-Bench with about 24% less quota and about 19% faster completion. The Carbon version is reported as Ultra-only.
How does CALL-E keep agent phone calls under the user’s control?
According to AI Rudder, the agent shows a call plan that the user must approve before dialing. A pre-call check blocks posing as emergency services or law enforcement. After each call, the user gets a summary, a confidence assessment and the supporting transcript lines.
I'm Maya — I write most of what you'll read here. I spent years as a copywriter before I got a little obsessed with what these AI tools can actually do, so now I spend my days poking at chatbots, breaking them, and writing up what's worth your time. Everything here is something I've actually tried. If a prompt didn't work for me, it doesn't make the cut.
Want to try any of this?
Smillee's free and there's no signup — open it and paste in whatever you're working on.
Start chatting →More from the blog
- Trends
AT&T Makes Its 500 Agents Earn Their Autonomy, AWS Sells Cited Answers as a Connector, and New Jersey Moves to Treat Chatbots as Products
AT&T says nearly 500 agentic workflows run in production, each supervised by a person until it reaches mid-to-high-90s accuracy. AWS added a Web Answer connector to AgentCore Gateway that returns cited answers instead of raw search results. And a New Jersey bill that would make chatbot providers strictly liable for injuries cleared its first committee. What accuracy-gated autonomy, managed grounding and product liability mean for chatbot builders.
- Trends
Claude Managed Agents Move Inside the Firewall, Gemini Enterprise Agents Sign In as You, and ChatGPT Retires Two Voice Modes
Anthropic opened a private beta that lets companies run the whole Claude Managed Agents service on their own infrastructure, Google made 3-legged OAuth generally available for A2A agents in Gemini Enterprise, and OpenAI set December 11 to retire Advanced and Standard voice in ChatGPT in favor of GPT-Live. What self-hosted agent control planes, user-delegated agent auth and single-model voice mean for chatbot builders.
- Trends
ChatGPT Wallet Lets Agents Pay With a Fingerprint, Gemini 4 Hydrogen Becomes the New Default, and Claude Joins the Meeting
OpenAI gave its dots agents a tokenized card wallet with a biometric check on every purchase, Google rolled out Gemini 4 Hydrogen across Search, the Gemini app and the API in place of Gemini 3.8 Flash, and Anthropic put Claude into Google Meet calls as a visible participant. What agent payments, silent default-model swaps and agents in live meetings mean for chatbot builders.