Small models do frontier work; platform layer races to match
Claude Haiku 4.5 and DoorDash delivery by Waymo, and The Deploy Log for this week. Every price and claim read off the publisher's own page, with a call on each.
Anthropic
Claude Haiku 4.5
DEPLOYif you run latency-sensitive coding or chat workloads
WHAT SHIPPEDClaude Haiku 4.5 is available today to all users at $1 per million input tokens and $5 per million output tokens, with coding performance similar to Claude Sonnet 4 at one third the cost and more than twice the speed.
THE CATCHHaiku 4.5 surpasses Sonnet 4 on computer use tasks and scored 65% accuracy on slide text generation versus 44% from a premium tier model, so the small model is not just cheaper but better at specific jobs.
IF YOU RUNClaude Code with multiple agent projects or rapid prototyping and want markedly more responsive coding
SOURCEAnthropic, Introducing Claude Haiku 4.5
Waymo
DoorDash delivery by Waymo
WATCHif you order DoorDash in Metro Phoenix
WHAT SHIPPEDWaymo and DoorDash are partnering to match DoorDash customers with fully autonomous Waymo vehicles for deliveries from DashMart across a 315 square mile service area in Phoenix, with more local merchants planned over time.
THE CATCHThe Waymo Driver has logged over 100 million autonomous miles and served over 10 million trips, but this delivery service starts only with DashMart orders and requires opting into autonomous vehicle delivery at checkout.
IF YOU RUNa DashPass membership and want the promotional Waymo ride hailing offer in Phoenix, San Francisco, or Los Angeles
SOURCEWaymo, Your DoorDash Order, Delivered by Waymo
The Deploy Log
Tools and products
| DEPLOYif you use Braintrust, Dash0, Statsig, or Kubiks Trace Drains on Marketplace. Vercel Drains now send traces and logs from projects to Marketplace observability providers including Braintrust, Dash0, Statsig, and Kubiks, with more providers coming soon. Vercel |
Left out
- Anomaly alerts public beta: a figure in the draft could not be found on the publisher's own page
- Braintrust on Vercel Marketplace: the condition it attaches to the call did not hold up
- On Demand Concurrent Builds: the condition it attaches to the call did not hold up
- Agent Skills: the condition it attaches to the call did not hold up
- NestJS zero configuration support: it did not carry a number that moved
- Claude Haiku 4.5 on AI Gateway: its own name for the thing was too long to print
The pattern
Claude Haiku 4.5 at $1 per million input tokens and $5 per million output tokens does frontier coding work at one third the cost of Claude Sonnet 4, while Waymo and DoorDash put a 315 square mile Phoenix delivery service on top of 100 million autonomous miles and 10 million trips, and Vercel Drains push traces and logs to Braintrust, Dash0, Statsig, and Kubiks. The small model beats a premium tier model on slide text generation with 65% accuracy versus 44%, the Waymo Driver serves only DashMart orders through an opt-in at checkout, and Vercel starts with four observability providers with more coming soon. Each launch is a platform layer racing to match a capability that already exists in a narrower form: Haiku 4.5 matches Sonnet 4 coding at lower cost, Waymo matches human delivery in one Phoenix area, and Vercel Drains match observability pipelines through a Marketplace.
The pattern is not that these systems are complete, but that they are specific and priced for adoption. Haiku 4.5 is more than twice the speed and better at computer use tasks, Waymo offers a promotional ride hailing offer in Phoenix, San Francisco, or Los Angeles for DashPass members, and Vercel Drains target projects that need logs and traces sent to third parties. Each item ships a small model doing frontier work in one domain while the platform layer around it races to match that work with distribution, pricing, or integrations.