DeepSeek Adds Standalone Search and Fetch APIs for Agents, Priced at 5 Yuan per 1,000 Calls
Standalone Search and Fetch APIs for agents, priced at 5 yuan per 1,000 calls.
AI Daily Recap
16 articles · 5 categories
The finishable daily brief
What happened in AI — Oct 17, 2026
Saturday, Oct 17, 2026
16 articles · 5 categories
read top to bottom · then stop
In 30 seconds
- DeepSeek launches Search and Fetch APIs for agents at 5 yuan (~$0.75) per 1,000 calls.
- DeepSeek V4.1-Flash-1017 reportedly beats larger rivals on agentic tasks at lower per-task cost.
- V4.1-Flash is tied more tightly to DeepSeek's own harness.
- Nono, Baa and Agentfork add kernel-level boundaries, process cleanup and step-level replay for agents.
- Reverse Engineer Anything links Claude Code and Codex to Ghidra, Hopper and IDA.
DeepSeek dominated the day: standalone Search and Fetch APIs at about $0.75 per 1,000 calls, plus the V4.1-Flash-1017 model aimed at cheaper agentic tasks.
Elsewhere, builders got new open tools for sandboxing, cleaning up, and replaying agents, and Meta's moves reignited talk of US open weights.
Agent Search and Fetch Become Priced APIs 3 items
DeepSeek shipped standalone Search and Fetch APIs for agents at 5 yuan (about $0.75) per 1,000 calls, giving builders a per-call price for web grounding.
DeepSeek prices web search for AI agents at $0.75 per 1,000 calls
Converts the price to $0.75 per 1,000 web-search calls, a number to compare against your current search provider.
DeepSeek moves up the stack: V4.1-Flash update ships alongside its own Search and Fetch APIs for agents
Ships alongside a V4.1-Flash update, moving DeepSeek from model supplier toward the agent tool layer.
DeepSeek V4.1-Flash-1017 Targets Agentic Cost 3 items
V4.1-Flash-1017 is pitched on agentic-task results and per-task cost, and is tied closely to DeepSeek's own harness.
Deepseek V4.1 Flash 1017 beats larger rivals in agentic tasks while once again massively cutting per-task costs
Reported to beat larger rivals on agentic tasks while cutting per-task cost again.
DeepSeek Ties Its New V4.1-Flash-1017 Model Tighter to Its Own Harness
Model is tied more tightly to DeepSeek's own harness, which matters if you run it through a different agent loop.
DeepSeek Harness v0.2.1-rc1 Arrives as an Unannounced V4.1 Flash Snapshot Briefly Surfaces
Harness v0.2.1-rc1 appeared as an unannounced V4.1 Flash snapshot briefly surfaced.
Open-Weight Landscape Shifts 2 items
Meta's frontier moves are being read as a return of US open weights, while small open models keep competing with larger tiers.
Meta shakes up the frontiers, open and closed
Interconnects frames Meta's latest moves across open and closed frontiers as the return of US open weights.
Dory 1.0 4B vs Qwen 3.8: a 4B against a three-tier line
Compares a 4B Dory 1.0 model against Qwen 3.8's three-tier line.
Agent Runtime Security and Debuggability 3 items
New open tools target the operational gaps around running agents: sandboxing, cleanup, and replay.
Nono: Agent runtime security. Kernel-enforced boundaries for AI agents
Nono enforces agent runtime boundaries at the kernel level.
Show HN: Baa – find and free what your AI agents left running on your Mac
Baa finds and frees processes AI agents left running on a Mac.
Show HN: Agentfork – Replay an LLM agent from any step and test fixes
Agentfork replays an LLM agent from any step so you can test a fix without rerunning the whole trace.
Coding Agents Meet Existing Tooling 2 items
Coding agents are being wired into mature tools such as Ghidra, IDA and fzf rather than replacing them.
Reverse Engineer Anything Project Brings AI Agents to Binary Analysis
Open-source Reverse Engineer Anything connects Claude Code and Codex to Ghidra, Hopper and IDA for binary analysis.
Using fzf with AI coding agents
Tips for using fzf alongside AI coding agents.
You are caught up for this edition