💲 LLM API Pricing
Chat per-token pricing plus image, video, speech, transcription and embedding rates — refreshed daily · updated 19h. Data: LiteLLM + OpenRouter, plus a small curated set (Kling, Hailuo, ElevenLabs, Cartesia) from official pricing pages.
All providersbedrock_converse 599bedrock_mantle 419fireworks_ai 328bedrock 275azure 211deepinfra 140novita 130gmi 122azure_ai 110openai 104vercel_ai_gateway 93empiriolabs 92kyma 83replicate 83
Cost calculator — what would your workload cost on each model?
| Model | Elo | Input $/M | Output $/M | Cache read | Cache write | Batch in/out | Context | History |
|---|---|---|---|---|---|---|---|---|
| meta.llama3-2-3b-instruct-v1:0bedrock | — | $0.17 | $0.17 | — | — | — | 128K | |
| mistral.mistral-7b-instruct-v0:2bedrock | — | $0.17 | $0.23 | — | — | — | 32K | |
| llama-4-scout-instructreplicate | — | $0.17 | $0.65 | — | — | — | — | |
| Qwen3.5-9B💬together_ai | — | $0.17 | $0.25 | — | — | — | 262K | |
| Qwen3-235B-A22B-FP8gmi | — | $0.17 | $1.09 | — | — | — | 33K | |
| deepseek-v4.1-flash💬prism | — | $0.17 | $0.63 | $0.0700 | — | — | 1.0M | |
| deepseek-v4-flash💬prism | 1432 | $0.17 | $0.21 | $0.0700 | — | — | 1.0M | |
| qwen3-8-27b💬empiriolabs | 1437 | $0.17 | $0.50 | $0.0800 | — | — | 262K | |
| qwen.qwen3-next-80b-a3b-instructbedrock_mantle | — | $0.17 | $1.44 | — | — | — | 256K | |
| mimo-v2.5💬novita | 1428 | $0.17 | $0.34 | $0.0034 | — | — | 1.0M | |
| qwen.qwen3-next-80b-a3b-instructbedrock_mantle | — | $0.17 | $1.44 | — | — | — | 256K | |
| qwen.qwen3-next-80b-a3b-instructbedrock_mantle | — | $0.17 | $1.44 | — | — | — | 256K | |
| qwen.qwen3-next-80b-a3b-instructbedrock_mantle | — | $0.17 | $1.44 | — | — | — | 256K | |
| qwen.qwen3-next-80b-a3b-instructbedrock_mantle | — | $0.17 | $1.44 | — | — | — | 256K | |
| qwen.qwen3-next-80b-a3b-instructbedrock_mantle | — | $0.17 | $1.44 | — | — | — | 256K | |
| google.gemma-4-31bbedrock_mantle | — | $0.17 | $0.48 | — | — | — | 256K | |
| qwen.qwen3-next-80b-a3b-instructbedrock_mantle | — | $0.17 | $1.44 | — | — | — | 256K | |
| qwen.qwen3-next-80b-a3b-instructbedrock_mantle | — | $0.17 | $1.44 | — | — | — | 256K | |
| google.gemma-4-31bbedrock_mantle | — | $0.17 | $0.48 | — | — | $0.0840 / $0.24 | 256K | |
| gpt-4o-mini-2024-07-18💬azure | 1286 | $0.17 | $0.66 | $0.0830 | — | $0.0830 / — | 128K | |
| gpt-4o-mini💬azure | — | $0.17 | $0.66 | $0.0830 | — | $0.0830 / — | 128K | |
| FW-GPT-OSS-120Bazure_ai | — | $0.17 | $0.66 | $0.0820 | — | — | 131K | |
| gpt-4o-mini💬azure | — | $0.17 | $0.66 | $0.0830 | — | $0.0830 / — | 128K | |
| gpt-4o-mini-2024-07-18💬azure | 1286 | $0.17 | $0.66 | $0.0830 | — | $0.0830 / — | 128K | |
| qwen3-6-flash:variant1empiriolabs | — | $0.17 | $0.99 | — | — | — | 984K | |
| qwen3.5-397b-a17b💬aihubmix | 1438 | $0.16 | $0.99 | — | — | — | 262K | |
| gemma-4-31bkyma | — | $0.16 | $0.47 | $0.16 | — | — | 128K | |
| gpt-6-luna💬kyma | — | $0.16 | $0.82 | $0.0135 | — | — | 1.1M | |
| qwen.qwen3-coder-30b-a3b-instructbedrock_mantle | — | $0.16 | $0.62 | — | — | — | 256K | |
| mistral.mistral-7b-instruct-v0:2bedrock | — | $0.16 | $0.22 | — | — | — | 32K | |
| mistral.ministral-3-3b-instructbedrock_mantle | — | $0.16 | $0.16 | — | — | — | 128K | |
| mistral.voxtral-small-24b-2507bedrock_mantle | — | $0.16 | $0.47 | — | — | — | 33K | |
| qwen.qwen3-32bbedrock_mantle | — | $0.16 | $0.62 | — | — | — | 32K | |
| qwen3-vl-32b-instruct👁️qwen_ai_platform | — | $0.16 | $0.64 | — | — | — | 131K | |
| meta.llama3-2-11b-instruct-v1:0bedrock | — | $0.16 | $0.16 | — | — | — | 128K | |
| openai.gpt-oss-120b-1:0bedrock_converse | — | $0.16 | $0.62 | — | — | — | 128K | |
| qwen3-vl-32b-thinkingqwencloud | — | $0.16 | $2.87 | — | — | — | 131K | |
| qwen3-vl-32b-instruct👁️qwencloud | — | $0.16 | $0.64 | — | — | — | 131K | |
| llama-3.2-11bvercel_ai_gateway | — | $0.16 | $0.16 | — | — | — | 128K | |
| mistral.ministral-3-3b-instructbedrock_converse | — | $0.16 | $0.16 | — | — | — | 128K | |
| us.meta.llama3-2-11b-instruct-v1:0bedrock | — | $0.16 | $0.16 | — | — | — | 128K | |
| openai.gpt-oss-120bbedrock_mantle | — | $0.16 | $0.62 | — | — | — | 131K | |
| mistral.voxtral-small-24b-2507bedrock | — | $0.16 | $0.47 | — | — | $0.0800 / $0.23 | 33K | |
| Qwen3.8-Flash💬gmi | — | $0.16 | $0.47 | $0.0160 | — | — | 1.0M | |
| qwen3-8-flash💬empiriolabs | — | $0.16 | $0.47 | — | — | — | 984K | |
| qwen3-vl-32b-thinkingdashscope | — | $0.16 | $2.87 | — | — | — | 131K | |
| qwen3-vl-32b-instruct👁️dashscope | — | $0.16 | $0.64 | — | — | — | 131K | |
| qwen3-vl-32b-thinkingqwen_ai_platform | — | $0.16 | $2.87 | — | — | — | 131K | |
| qwen.qwen3-32b-v1:0bedrock_converse | — | $0.16 | $0.62 | — | — | — | 33K | |
| hy3💬kyma | 1439 | $0.16 | $0.66 | — | — | — | 262K |
Recent price changes
- ▼ 1.4%Z.ai: GLM 5.3 Flash (batch)z-ai · output $0.0690 → $0.068014h
- ▼ 4.2%Z.ai: GLM 5.3 Flash (batch)z-ai · input $0.0240 → $0.023014h
- ▼ 84.3%Z.ai: GLM 5.3 Flashz-ai · output $0.50 → $0.078414h
- ▼ 73.4%Z.ai: GLM 5.3 Flashz-ai · input $0.15 → $0.039914h
- ▼ 64.4%Z.ai: GLM 5.2z-ai · output $6.80 → $2.4214h
- ▲ 8757.1%Z.ai: GLM 5.2z-ai · input $0.0070 → $0.6214h
- ▲ 66.7%Qwen: Qwen3.5-27Bqwen · output $1.56 → $2.6014h
- ▲ 33.3%Qwen: Qwen3.5-27Bqwen · input $0.20 → $0.2614h
- ▲ 16.7%Particle AI: Particle-1particle-ai · input $0.0600 → $0.070014h
- ▼ 40.0%OpenAI: gpt-oss-120bopenai · output $0.60 → $0.3614h
- ▼ 40.0%OpenAI: gpt-oss-120bopenai · input $0.15 → $0.090014h
- ▼ 1.0%MoonshotAI: Kimi Latestmoonshotai · input $0.18 → $0.1814h
- ▼ 14.3%MoonshotAI: Kimi K2.7 Codemoonshotai · output $3.50 → $3.0014h
- ▼ 1.5%Meta: Llama 3.2 3B Instructmeta-llama · output $0.34 → $0.3314h
- ▼ 1.8%Meta: Llama 3.2 3B Instructmeta-llama · input $0.0509 → $0.050014h
- ▼ 10.8%DeepSeek: DeepSeek V4 Pro 0813deepseek · output $4.44 → $3.9614h
- ▲ 1089.2%DeepSeek: DeepSeek V4 Pro 0813deepseek · input $0.11 → $1.3214h
- ▼ 40.0%DeepSeek: DeepSeek V4 Flash Latestdeepseek · output $0.30 → $0.1814h
- ▲ 20.0%DeepSeek: DeepSeek V4 Flash Latestdeepseek · input $0.0500 → $0.060014h
- ▼ 23.3%DeepSeek: DeepSeek V4 Flash 0731deepseek · output $0.30 → $0.2314h
Price cuts are green — they make AI cheaper for everyone.