Community Discussion · Policy

Honda Integrates In-Car LLMs: Balancing Cost and Experience at the Critical Point

Lao FanJul 152026/07/15 66 views

Last week we were still discussing the bottlenecks of deploying large models on the edge in cockpit systems, and today I saw Honda America announce they're integrating Google Gemini into mainstream models like the Civic, Accord, and CR-V. As an engineer who deals with vehicle costs and electronic control systems year-round, my first reaction wasn't "how much has the intelligence level improved," but rather "what power consumption does this chip have to handle, and can the vehicle's overall electrical balance and thermal management cope?"

Policy

2 replies

Ctrl + Enter to reply

Engineer XueJul 27(edited)

[quote="fan_xiujie, post:1, topic:725"]

Last week we were still discussing the bottlenecks of deploying edge-side large models in cockpits, and today I saw Honda America announce integrating Google Gemini into mainstream models like the Civic, Accord, and CR-V. As an engineer who deals with vehicle costs and electronic control systems year-round, my first reaction wasn't "how much did the intelligence level improve," but "how high power consumption will this chip handle, and can the vehicle's electrical balance and thermal management cope?"

Let's set up a practical scenario. Suppose you're driving a CR-V integrated with Gemini, wanting to adjust navigation, ask about charging stations along the route, and simultaneously lower the AC by two degrees on the highway—the traditional…

[/quote]

I've tried several lightweight models. Intent accuracy does drop by 10-15%, but cockpit scenarios don't actually require such high generalization capabilities. If limited to domains like navigation and AC, after pruning and quantization, power consumption can be kept under 2W—it just depends on whether Honda is willing to do customized development.

TiangongJul 26(edited)

[quote="fan_xiujie, post:1, topic:725"]

Just last week we were discussing the bottlenecks of deploying large models on edge devices in cockpits, and today Honda US announced they would integrate Google Gemini into mainstream models like the Civic, Accord, and CR-V. As an engineer who deals with whole-vehicle costs and electronic control systems daily, my first reaction wasn't "how much has intelligence improved," but rather "how much power consumption will this chip handle, and can the vehicle's electrical balance and thermal management cope?"

Let's look at a real scenario. Suppose you're driving a CR-V integrated with Gemini on the highway, wanting to adjust navigation, ask about charging stations along the route, and simultaneously lower the AC by two degrees—traditional…

[/quote]

Honda's move shows they're betting on the cost-reduction curve of edge large models. But continuous power consumption of 3-5W for a car manufacturer selling a million units annually means heat dissipation and electrical balance engineering amortization might be more headache-inducing than chip costs. Have you calculated how much intent understanding accuracy would be sacrificed if switching to a lower-power lightweight model?