Claude 4 API: Tool Use Streaming & Prompt Caching vs Cohere Command R3

“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: a hosted RAG-optimized language model with a first-class fine-tuning API you can actually call without a sales call. The DX bet is that self-serve fine-tuning lowers the activation energy for enterprise customization — and that's the right bet. The 128K window is table stakes at this point, but the multilingual grounding improvements are where Cohere has actually done real work rather than just scaling context. The moment of truth is whether the fine-tuning API docs are good enough to onboard without hand-holding — if it's one endpoint with a clear schema and a sensible job-polling pattern, this earns the ship. The specific decision that works here is putting fine-tuning behind an API instead of a wizard, which means it composes into deployment pipelines.”

“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.”

“Category is enterprise LLM API, direct competitors are OpenAI GPT-4o, Anthropic Claude 3.5, and Google Gemini 1.5 Pro — all of whom have 128K+ context windows and fine-tuning options. Cohere's actual differentiator is enterprise deployment posture: on-prem, private cloud, and data residency options that OpenAI still can't match for regulated industries. This breaks when a Fortune 500 IT department discovers the fine-tuning API doesn't yet support their private VPC deployment, which is precisely the customer Cohere is targeting. What kills this in 12 months is not a competitor — it's Cohere's own pricing as fine-tuning compute costs hit enterprise budgets that expected SaaS not metered AI. To be wrong about the ship: the team would have to fail to close the gap between self-serve and enterprise contract customers before the burn rate forces a pivot.”

“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 is falsifiable: enterprise teams will converge on fine-tuned, domain-specific RAG models rather than prompt-engineering general models, and they'll want to own that customization loop without vendor mediation. That thesis requires that fine-tuning costs keep falling faster than general model capability keeps rising — if GPT-5 class models make fine-tuning unnecessary for most enterprise tasks, Command R3's differentiation collapses. The second-order effect if this works is structural: self-serve fine-tuning APIs turn enterprise AI customization into a DevOps problem rather than an AI research problem, which shifts power from AI consultancies to internal platform teams. Cohere is on-time to the trend of enterprise model customization — not early, not late — but the multilingual angle on 23 languages is genuinely early to a market where most competitors are still English-first. The future state where this is infrastructure: every regulated-industry RAG pipeline has a Cohere fine-tuned model at its core the same way they have a Snowflake data warehouse.”

“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 is a VP of Engineering or AI platform lead at a mid-market to enterprise company who has already approved a RAG budget and needs a model that won't leak their data to a competitor's training pipeline — that's a real budget line and Cohere owns it more credibly than OpenAI. The self-serve fine-tuning API is a smart pricing unlock: it moves customization from a six-figure enterprise conversation to a metered API call, which compresses the sales cycle and creates natural expansion revenue as teams fine-tune more models. The moat is not the model quality — it's the data residency and compliance posture that Cohere has built over years, which takes time to replicate. The stress test that concerns me: if Azure OpenAI closes the compliance gap further, Cohere's addressable market shrinks to the subset that truly cannot use US hyperscalers, which is real but not massive.”