Claude 4 API: Tool Use Streaming & Prompt Caching vs Perplexity Deep Research API

“The primitive here is clean: incremental tool-call deltas over SSE, and a cache-control header you attach to prompt segments to pin them server-side. The DX bet is that complexity lives in the HTTP layer, not in a new SDK abstraction — you opt in per-request, no new mental model required. The moment of truth is calling `stream=true` on a tool-use request and watching partial JSON arguments arrive before the model finishes thinking, which actually matters for agent loops where you want to dispatch work early. This is not a weekend-script replacement — implementing correct incremental JSON parsing for partial tool arguments plus a reliable distributed cache with 2M token capacity is a real engineering problem Anthropic has solved for you. The specific decision that earns the ship: cache invalidation is explicit and cache hits are reflected in the usage object, so you can actually measure what you're saving instead of guessing.”

“The primitive here is clean: POST a research question, get back a structured report with citations — no orchestration layer required, no managing a scraping fleet, no stitching together search APIs. The DX bet is that complexity lives entirely inside the endpoint, which is the right call for most integration scenarios. The moment of truth is whether the output schema is stable and documented well enough to build against without treating every response as freeform text, and Perplexity's track record on API consistency is decent if not exceptional. This isn't something you'd replicate in a weekend — the multi-step planning and source arbitration is genuinely non-trivial — but the free tier being available for prototyping is the thing that actually earns the ship here.”

“Direct competitors are OpenAI's cached completions and Google's context caching in Gemini 1.5 — both shipping for months — so Anthropic is catching up, not leading. The specific scenario where this breaks: cache hit rates depend entirely on prompt structure, and developers who dynamically compose system prompts (inserting user-specific context at the top) will see near-zero cache utilization and pay full price while assuming they're saving money. The prediction: this feature doesn't get killed — it becomes table stakes infrastructure and Anthropic wins by having the largest cache window (2M vs. competitors' current limits). What would have to be true for me to be wrong: OpenAI ships a 10M token cache window before Anthropic's ecosystem matures, commoditizing the advantage. Still a ship because the streaming tool-use delta is genuinely differentiated — no competitor has clean partial-argument streaming for tool calls yet, and that changes agent loop architecture in ways that matter.”

“Direct competitor is Exa's research endpoint combined with a Claude or GPT synthesis call — and yes, you can stitch that together yourself, but Perplexity has a genuine edge in real-time web indexing depth that raw Exa plus LLM doesn't fully replicate yet. The scenario where this breaks is high-frequency programmatic research at scale: session-token pricing with 'contact for volume' is a wall that will hit enterprise devs exactly when they're most committed to the integration. What kills this in 12 months isn't a competitor — it's OpenAI or Google shipping a native deep research endpoint at commodity pricing, which both companies have every incentive to do given their existing search infrastructure. Ship now, but build your abstraction layer thin so you can swap providers.”

“The thesis this bets on: by 2027, the dominant AI application architecture is a persistent agent with a large, stable context (tools, memory, instructions) that gets reused across thousands of user interactions — making context I/O cost the primary unit economics lever, not generation cost. The dependency that has to hold: agents don't collapse back to stateless chatbots, and context windows keep growing faster than per-token prices fall. The second-order effect nobody's talking about: prompt caching at 2M tokens makes it economically viable to give every enterprise user a fully-loaded, role-specific agent context at request time — which shifts competitive differentiation from 'who has the best model' to 'who has the best cached context corpus,' effectively making knowledge curation the new moat. This tool is riding the trend of context-window expansion-as-infrastructure, and it's on-time, not early — but the streaming tool-use primitive is ahead of the curve on agent loop efficiency. The future state where this is infrastructure: every production agentic system has a cache manifest the same way it has a CDN config.”

“The thesis here is falsifiable: within three years, research as a discrete cognitive task gets fully externalized into API calls, and every knowledge-worker application has a 'go find out' endpoint the same way every e-commerce application has a payment endpoint today. What has to go right is that output quality crosses the trust threshold for professional use cases — legal, financial, strategy — which requires both accuracy gains and citation provenance robust enough to audit. The second-order effect if this wins is that the research analyst role gets restructured around output validation and prompt strategy rather than raw information gathering, which shifts power toward developers who own the integration layer. Perplexity is genuinely early on this specific primitive — the trend toward externalizing reasoning steps into APIs is real and accelerating, and they're positioned as infrastructure rather than application, which is where you want to be.”

“The buyer is the engineering team at any company running Claude in production with long system prompts or multi-step agents — this comes out of the AI infrastructure budget, not a new budget line, which means no procurement friction. The pricing architecture is sound: cache reads at ~90% discount means the savings are real and measurable in the first billing cycle, which creates immediate retention — developers who restructure prompts to maximize cache hits are now architecturally coupled to Anthropic's caching implementation. The moat question is the honest one: this is infrastructure that OpenAI and Google will match, so the defensible position isn't the feature itself but the ecosystem of developers who've restructured their codebases around it. What survives a 10x model price drop: the streaming tool-use architecture, because that's about latency, not cost. The specific business decision that makes this viable is pricing cache reads as a separate SKU — it lets Anthropic capture value from high-volume production workloads without losing price-sensitive experimenters.”

“The buyer here is an enterprise developer with a research automation budget, which is a real buyer with a real budget — so credit for that. The problem is 'contact for volume' pricing on the thing developers will use at scale is a conversion killer; by the time a team has prototyped on the free tier and needs to talk to sales, half of them have already evaluated the DIY path. The moat is thin: Perplexity's advantage is their index freshness and citation quality, but Google's Gemini with Grounding and OpenAI's search integration are closing that gap every quarter with distribution advantages Perplexity cannot match. This is a good product in search of a business model that can survive the next 18 months of platform competition.”