Spending $2.5k/month on Sonnet/Opus — worth switching more to GPT-5.5/Codex?
Currently I’m spending around $2k–$2.5k/month on Sonnet 4.6, with about 5% Opus 4.7 usage, mainly for development and business management through the Anthropic API.
Honestly, I’m extremely happy with the results for the money.
But I’m also trying to optimize costs and see whether I can get similar quality for less.
What I’ve done so far:
- Offloaded most coding work to GPT-5.5 using the $200 OpenAI plan
- For coding, it’s been great
- I also tried DeepSeek V4 Pro, but it felt too slow for my taste
- GPT-5.5 also feels a bit slower than Sonnet/Opus, but with multiple tabs open it works fine
Now I’m considering shifting more toward Codex/GPT to reduce my Anthropic API costs and get more value from the $200 plan.
But I keep hearing that Codex/GPT isn’t as good for:
- answering customer emails
- writing knowledge base articles
- handling employee planning/operations
- general business communication
And that Anthropic models (especially Sonnet/Opus) are still better for those tasks.
So I’m curious what others are doing:
- Are you running everything on Codex/GPT now?
- Or are you still using Anthropic for writing/business tasks and Codex mainly for development?
- Any setups/workflows/models you’d recommend for optimizing quality vs cost?
Would love to hear how others are structuring their stack.
UPDATE:
Really appreciate all the input from you guys!
We’ve currently offloaded all development work to GPT-5.5 through OAuth, and we’re testing DeepSeek V4 Flash as a replacement for Sonnet.
So far GPT-5.5 has been great for coding workloads, especially when paired with multiple parallel chats/agents.