Can you share more about what makes them poor at coding? I haven't tried Google models, but I'm curious what issues you hit?
Yeah, having review/orchestration + subagents seems like the stndard way to optimize for cost/performnce.
Using OpenCode has been on the list to try. Logically, the value of a bundle from a larger company should technically beat out a smaller purveyor of open-weight models right? I also wonder how much OpenCode's token/sec throttling will impact my productivity?
I've heard that folks using using a mid-tier Anthropic model like Sonnet 5.5 easily run into limits. When I use GitHub Copilot (for work) with Anthropic models I can easily rack up quite a significant bill.
This is why I'm comparing value for the money.
While Google may not have the "best" models, what's the point if you hit a wall with Claude Code? The extra baseline quota with Anti-gravity could be worth it.
The Google models are great "general purpose" and research LLMs, but they're relatively poor at coding.
The way I see it, you have two options.
1) Get on the Claude Max 20x plan and use Opus 5.5 for review/orchestration and Haiku 5.5 for grunt work. Haiku 5.5 is decent, cheap, and fast.
2) Deepseek 4.1 Flash in OpenCode.
You should also test MiMo v2.6. It's a little bit uneven, but might work well for whatever you're trying to do, and it's also cheap.
Can you share more about what makes them poor at coding? I haven't tried Google models, but I'm curious what issues you hit?
Yeah, having review/orchestration + subagents seems like the stndard way to optimize for cost/performnce.
Using OpenCode has been on the list to try. Logically, the value of a bundle from a larger company should technically beat out a smaller purveyor of open-weight models right? I also wonder how much OpenCode's token/sec throttling will impact my productivity?
Honestly. Just buy the Claude Prox 5x plan. it works great.
I've heard that folks using using a mid-tier Anthropic model like Sonnet 5.5 easily run into limits. When I use GitHub Copilot (for work) with Anthropic models I can easily rack up quite a significant bill.
This is why I'm comparing value for the money.
While Google may not have the "best" models, what's the point if you hit a wall with Claude Code? The extra baseline quota with Anti-gravity could be worth it.