Early-access prices are provider rates with no ModelRelay markup. Intelligence by Artificial Analysis.
| Anthropic | Claude Fable 5 | 2.19s — evidence: controlled · n=8 | 119.3tps — evidence: controlled · n=8 | $10.00 | $50.00 | 62.1 | |
| Anthropic | Claude Opus 5 | 1.02s — evidence: controlled · n=8 | 115.2tps — evidence: controlled · n=8 | $5.00 | $25.00 | 63.1 | |
| Anthropic | Claude Sonnet 5 | 0.69s — evidence: controlled · n=8 | 152.0tps — evidence: controlled · n=8 | $2.00 | $10.00 | 55.3 | |
| Google AI Studio | Gemini 3.7 Flash | 0.78s — evidence: controlled · n=8 | 613.4tps — evidence: controlled · n=8 | $0.75 | $3.75 | 56.0 | |
| OpenAI | GPT-5.6 Luna | 0.77s — evidence: controlled · n=8 | 231.4tps — evidence: controlled · n=8 | $0.20 ≤272K input | $1.20 ≤272K input | 52.3 | |
| OpenAI | GPT-5.6 Sol | 1.52s — evidence: controlled · n=8 | 121.4tps — evidence: controlled · n=8 | $5.00 ≤272K input | $30.00 ≤272K input | 60.9 | |
| OpenAI | GPT-5.6 Terra | 0.66s — evidence: controlled · n=8 | 186.3tps — evidence: controlled · n=8 | $2.00 ≤272K input | $12.00 ≤272K input | 56.6 | |
| xAI | Grok 4.6 | 1.40s — evidence: controlled · n=8 | 103.6tps — evidence: controlled · n=8 | $2.00 ≤200K input | $6.00 ≤200K input | 60.9 |
TTFT is time to first visible answer token; TPS is visible answer tokens per second after that token. Both use a fixed prompt at each model's minimum reasoning. Speed combines both phases into time to a 500-token answer (TTFT + 500/TPS); less time ranks higher, and neither column changes meaning. Intelligence is the Artificial Analysis Intelligence Index at each model's highest scored variant.
Use ModelRelay's Responses API directly, or point an existing OpenAI or Anthropic SDK at the compatible endpoints.
curl https://api.modelrelay.ai/api/v1/responses \
-H "Authorization: Bearer $MODELRELAY_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"claude-sonnet-5","input":"Hello"}'Change the model. Keep everything else.
ModelRelay is invite-only while we onboard early teams.