Tegy model cost switch
Replacing Qwen with Claude is mostly a 3.5x to 17.4x cost move.
This estimates the same token volume currently served by Qwen3.7 Plus on OpenRouter, priced against Anthropic's current Claude API models.
Token prices
Qwen3.7 Plus via OpenRouter
$0.32 / $1.28 input / output MTok
Claude Haiku 4.5
$1 / $5 input / output MTok
Claude Sonnet 4.6
$3 / $15 input / output MTok
Claude Opus 4.8
$5 / $25 input / output MTok
Same traffic, same mix
If the workload is output-heavy, the Anthropic premium rises slightly. If repeated context is cacheable, Anthropic prompt caching can materially reduce the input side.
Formula
Qwen = input_M * 0.32 + output_M * 1.28
Haiku = input_M * 1 + output_M * 5
Sonnet = input_M * 3 + output_M * 15
Opus = input_M * 5 + output_M * 25
Haiku = input_M * 1 + output_M * 5
Sonnet = input_M * 3 + output_M * 15
Opus = input_M * 5 + output_M * 25
Exact totals require the OpenRouter or Cloudflare AI Gateway usage export split by input and output tokens.
Budget examples
| Qwen spend | Haiku | Sonnet | Opus |
|---|---|---|---|
| $10 | $34.72 | $104.17 | $173.61 |
| $25 | $86.81 | $260.42 | $434.03 |
| $50 | $173.61 | $520.83 | $868.06 |
Decision read
Haiku
Closest Anthropic cost option. Still roughly 3.5x Qwen for this agent-style mix.
Sonnet
Likely quality upgrade path, but budget should assume about 10x the same Qwen traffic.
Opus
Reserve for escalation, evals, or high-value turns unless caching changes the economics.