The high-throughput API designed for AI engineers. Stream tokens with sub-40ms time-to-first-token, guaranteed uptime SLAs, and drop-in SDK compatibility.
import { InferenceClient } from '@getbolt/ai-engine';
const ai = new InferenceClient({ apiKey: process.env.BOLT_AI_KEY });
const stream = await ai.chat.stream({
model: 'bolt-fast-70b-instruct',
messages: [{ role: 'user', content: 'Generate high-performance Postgres index config' }],
response_format: { type: 'json_object' },
max_tokens: 1024,
});
for await (const chunk of stream) {
process.stdout.write(chunk.content);
}Independent benchmarks measured on real streaming user requests.
| Model Architecture | Throughput (Tokens/s) | Time to First Token (TTFT) | Pricing / 1M Tokens |
|---|---|---|---|
| Bolt Llama-3.3 70B | 310 tok/s | 36 ms | $0.59 / $0.79 |
| Bolt DeepSeek R1 Distill | 265 tok/s | 44 ms | $0.45 / $0.65 |
| Bolt Qwen 2.5 Coder 32B | 380 tok/s | 28 ms | $0.35 / $0.48 |
Predictable token pricing with generous free tiers for all developers.
Build and test prototype applications
For live SaaS apps requiring high concurrency
Reserved H100/H200 nodes for enterprise
Generate your secret API key in 30 seconds and test in your terminal.
Get Your Free API Key Now