Free
$0
- Models
- Selected
- TPM
- 160,000
- Concurrency
- Shared capacity
- Queue
- May queue
- SLA
- No
LLM API pricing
USD per 1M tokens. Compare input, cache read, cache write and output prices.
Catalog updated: 2026-10-19.
| Model | Input | Cache Read | Cache Write | Output |
|---|---|---|---|---|
 Qwen3.8-27B | $0 | $0 | — | $0 |
| Qwen3.8-2.4T-A95B | $1.95 | $0.25 | — | $5.7 |
 GLM-5.3-Flash | $0.08 | $0.023 | — | $0.28 |
 Kimi K2.7 Code | $0.69 | $0.15 | — | $3.29 |
 DeepSeek V4 Pro | $0.79 | $0.026 | — | $2.38 |
| GLM-5.3 | $0.82 | $0.2 | — | $2.88 |
| Kimi K3 | $2.35 | $0.23 | — | $11.76 |
 MiniMax M3 | $0.3 | $0.05 | — | $1.19 |
| DeepSeek V4.1 Flash | $0.17 | $0.005 | — | $0.7 |
| DeepSeek V4 Flash 0731Legacy compatibility alias | $0.17 | $0.005 | — | $0.7 |
| Qwen3.8-Flash | $0.13 | $0.02 | — | $0.42 |
| GLM-5.2 | $0.53 | $0.12 | — | $1.72 |
| GLM-5.1 | $1.18 | $0.12 | — | $4.17 |
| GLM-5 | $0.89 | $0.11 | — | $3.14 |
| Kimi-K2.6 | $0.87 | $0.07 | — | $3.51 |
| Kimi-K2.5 | $0.54 | $0.06 | — | $2.78 |
| MiniMax-M2.7 | $0.38 | $0.03 | — | $1.48 |
| MiniMax-M2.5 | $0.27 | $0.03 | — | $1.03 |
 Claude Sonnet 5.5 | $1 | $0.05 | $1.25 | $5 |
 Gemini Nano Banana 2.1 | $1.2 | $0.12 | — | $6 |
 GPT Image 2.5 Sunburst | $4 | $1 | — | $0 |
| GPT Image 2.5 Flare | $4 | $1 | — | $0 |
| Gemini 4 Argon | $1.6 | $0.08 | — | $8 |
| Claude Opus 5.5 | $2 | $0.1 | $2.5 | $10 |
| Claude Haiku 5.5 | $0.05 | $0.005 | $0.0625 | $0.25 |
| Claude Mythos 5.1 | $5 | $0.125 | $6.25 | $25 |
| Gemini 3.8 Flash | $0.6 | $0.06 | — | $3 |
| Claude Fable 5.1 | $5 | $0.125 | $6.25 | $25 |
| GPT-6 Luna | $0.08 | $0.008 | $0.1 | $0.4 |
| GPT-6.1 Sol | $1.6 | $0.08 | $2 | $8 |
| GPT-6 Astra | $8 | $0.8 | $10 | $40 |
These are text-token rates from the public catalog. Image generation is billed separately; a zero text-output rate does not mean free images. Confirm image pricing and endpoint availability in the Console before use.
Qwen3.8-27B input, cached input and output are free. See the Free Models page for availability and limits.
Try Free →Up to 30M free tokens every month. No credit card.
A one-time balance top-up unlocks more Qwen capacity. $10 and $50 are cumulative recharge thresholds, not monthly fees. Qwen3.8-27B stays $0 per token in every tier.
| Tier | Cumulative net recharge | Monthly tokens | Context | TPM |
|---|---|---|---|---|
| Free | No recharge required | 30M | 128K | 160,000 |
| Starter | $10 or more | 100M | 128K | 300,000 |
| Pro | $50 or more | 300M | 256K | 600,000 |
Free accounts may queue during busy periods. Recharged accounts receive queue priority; capacity is shared and no latency SLA is offered.
Monthly allowances count input and output, including reasoning, across all API keys on your account. Context includes input and requested output; output is capped at 8,192 tokens. There is no daily quota. Allowances reset on the first of each month at 00:00 UTC (08:00 Beijing time); unused tokens do not roll over. Requests reserve their maximum token usage before running, then settle to actual usage. No automatic paid overage: exhausted monthly quota returns 429; wait for reset or unlock a higher tier.
Tiers use cumulative completed USD recharge principal, less refunds, excluding payment fees and promotional grants. Your recharge remains spendable API balance; using that balance does not lower your tier. Upgrades keep usage already consumed this month. Refunds may change your tier.
Serverless
Per token
- Models
- Paid models
- TPM
- Standard
- Concurrency
- Higher
- Queue
- Priority
- SLA
- Per agreement
Reserved
Committed capacity
- Models
- Selected
- TPM
- Guaranteed
- Concurrency
- Guaranteed
- Queue
- Reserved
- SLA
- Contract
Dedicated
Custom
- Models
- Custom
- TPM
- Dedicated
- Concurrency
- Custom
- Queue
- Isolated
- SLA
- Contract
Any SLA for paid Serverless usage is defined by the applicable service agreement. It does not cover Qwen3.8-27B on any Free, Starter or Pro tier. Qwen capacity is shared, subject to fair-use rate limits, with no SLA or reserved capacity. $10 / $50 top-ups raise Qwen allowance, TPM and context limits; they do not guarantee production performance.