Summary
- SayGM prices Claude about 11 percent below Anthropic list and about 16 percent below OpenRouter.
- OpenRouter is the most expensive provider: list price plus a 5.5 percent top-up fee.
- Haiku 4.5 costs half of Sonnet 5, which costs two fifths of Opus 5.
- Output tokens are 56 percent of a Claude bill at an 80/20 mix.
- Prompt caching only pays while a conversation stays on one provider.
01
Three providers of Claude
| Provider | Price | API | Fallback |
|---|---|---|---|
| Anthropic direct | List | Native Anthropic | None across providers |
| OpenRouter | List plus 5.5% fee | OpenAI-compatible | Retry on next provider |
| SayGM | At or below list | Native Anthropic | Cascade models |
Only one of the three costs less than going direct. The other trade-offs are API shape and what happens when a provider fails.
02
Prices side by side
| Model | Anthropic list | OpenRouter | SayGM |
|---|---|---|---|
| Claude Opus 5 | $5.00 / $25.00 | $5.28 / $26.38 | $4.45 / $22.23 |
| Claude Sonnet 5 | $2.00 / $10.00 | $2.11 / $10.55 | $1.78 / $8.89 |
| Claude Haiku 4.5 | $1.00 / $5.00 | $1.06 / $5.28 | $0.89 / $4.45 |
- Provider direct, list price
- OpenRouter, fee included
- SayGM
Show data as a table
| Provider direct, list price | OpenRouter, fee included | SayGM | |
|---|---|---|---|
| Claude Fable 5.1 | $1,250 | $1,319 | $1,111 |
| Claude Opus 5.5 | $500 | $528 | $445 |
| Claude Sonnet 5.5 | $250 | $264 | $222 |
| Claude Opus 5 | $625 | $660 | $556 |
| Claude Sonnet 5 | $250 | $264 | $222 |
| Claude Haiku 4.5 | $125 | $132 | $111 |
For Sonnet 5 at 50M input and 15M output tokens a month: $250 direct, $264 on OpenRouter, $222 on SayGM. The ordering is the same for every Claude model.
03
Model choice matters more than provider
On the same workload, moving from Opus 5 to Sonnet 5 saves $334 a month on SayGM. Moving Sonnet 5 from OpenRouter to SayGM saves $41. Pick the model first, then the route.
| Model | Monthly cost | Best for |
|---|---|---|
| Claude Opus 5 | $556 | Agentic coding, long multi-step tasks |
| Claude Sonnet 5 | $222 | Daily coding in editors and agents |
| Claude Haiku 4.5 | $111 | Classification, extraction, pipelines |
See Opus 5 vs Sonnet 5 for the full breakdown.
04
Output is most of the bill
Every Claude model prices output at five times input. At an 80/20 mix that makes output 56 percent of the cost. One saved output token is worth five saved input tokens.
- Input tokens (80% of volume)
- Output tokens (20% of volume)
Show data as a table
| Model | Share of bill from input | Share of bill from output |
|---|---|---|
| Claude Opus 5 | 44% | 56% |
| Claude Sonnet 5 | 44% | 56% |
| Claude Haiku 4.5 | 44% | 56% |
05
Keep prompt caching working
Caching is where Claude costs are won or lost on long system prompts and agent loops. A cached prefix only hits if the next request reaches the same provider. SayGM keeps a conversation on one provider. Gateways that spread a conversation across providers silently lose the discount.
06
Route easy work to Haiku
Send simple requests to Haiku 4.5 and escalate to Sonnet 5 on failure. If escalated requests pay for both attempts, the cascade saves money while fewer than half of requests escalate. At 30 percent escalation the Sonnet workload above drops from $222 to $178 on SayGM.
SayGM's cascade models do this at the gateway level. The cost levers guide works through the arithmetic.
07
Recommendation
Prices in this guide are a October 2026 snapshot and move each epoch.
Current SayGM pricing (opens in a new tab)