Home › Best › cheapest LLM to run The cheapest open LLMs to run via your own API key Open LLMs ranked by provider-published $/1M-token rates captured in Spanvero's dated table — cheapest first, $0 markup. Size-based estimates are excluded from the ordering, and a retired upstream rate is labeled last-known. Local and rented-GPU routes are shown on each model page.
How this is ranked: Built only from provider-sourced rates captured in the dated price table; size-based estimates are excluded from this ranking. 'Cheapest' describes the published blended token rate, not model quality or the final cost of a workload.
1. Ling 2.6 flash — inclusionAI, 107.5B · $0.02/1M API · ~81 GB VRAM · commercial OK2. Ling 3.0 flash — inclusionAI, 127.5B · $0.04/1M API · ~93 GB VRAM · commercial OK3. Gemma 2 9B Instruct — Google, 9B · $0.06/1M API · ~9.0 GB VRAM · commercial OK4. Nex N2 mini — nex-agi, 35.1B · $0.06/1M API · ~41 GB VRAM · commercial OK5. Llama 3.1 8B Instruct — Meta, 8B · $0.07/1M API · ~8.0 GB VRAM · commercial OK6. Mistral Small 3 (24B, 2501) — Mistral AI, 23.6B · $0.07/1M API · ~20 GB VRAM · commercial OK7. granite 4.1 8b — ibm-granite, 8.8B · $0.08/1M API · ~9.0 GB VRAM · commercial OK8. gpt-oss-20b — OpenAI, 21B · $0.08/1M API · ~15 GB VRAM · commercial OK9. DeepSeek V4 Flash — deepseek-ai, 158.1B · $0.09/1M API · ~107 GB VRAM · commercial OK10. Laguna XS 2.1 — poolside, 33.4B · $0.09/1M API · ~24 GB VRAM · non-commercial11. Trinity Mini — arcee-ai, 26.1B · $0.10/1M API · ~19 GB VRAM · non-commercial12. Gemma 3 12B — Google, 12B · $0.10/1M API · ~13 GB VRAM · commercial OK13. gpt-oss-120b — OpenAI, 117B · $0.10/1M API · ~80 GB VRAM · commercial OK14. Phi-4 — Microsoft, 14B · $0.11/1M API · ~13 GB VRAM · commercial OK15. Llama 3.2 1B Instruct — Meta, 1.2B · $0.11/1M API · ~3.0 GB VRAM · commercial OK16. Qwen3 30B A3B Instruct 2507 — Qwen, 30.5B · $0.12/1M API · ~22 GB VRAM · commercial OK17. Laguna S 2.1 — poolside, 117.6B · $0.14/1M API · ~81 GB VRAM · non-commercial18. Qwen2.5 7B Instruct — Alibaba, 7B · $0.15/1M API · ~7.0 GB VRAM · commercial OK19. Laguna XS.2 — poolside, 33.4B · $0.15/1M API · ~24 GB VRAM · commercial OK20. Qwen3 Coder 30B A3B Instruct — Qwen, 30.5B · $0.18/1M API · ~22 GB VRAM · commercial OK21. Qwen3 14B — Qwen, 14.8B · $0.18/1M API · ~13 GB VRAM · commercial OK22. Qwen3-32B — Alibaba, 32.8B · $0.18/1M API · ~25 GB VRAM · commercial OK23. gpt oss safeguard 20b — openai, 21.5B · $0.19/1M API · ~16 GB VRAM · commercial OK24. Llama 3.2 3B Instruct — Meta, 3B · $0.19/1M API · ~5.0 GB VRAM · commercial OK25. Mistral 7B Instruct v0.3 — Mistral AI, 7.2B · $0.20/1M API · ~8.0 GB VRAM · commercial OK26. Llama 4 Scout (17B-16E) — Meta, 109B · $0.20/1M API · ~77 GB VRAM · commercial OK27. Llama 3.3 70B Instruct — Meta, 70B · $0.21/1M API · ~53 GB VRAM · commercial OK28. DeepSeek V4 Flash 0731 — deepseek-ai, 304.2B · $0.21/1M API · ~205 GB VRAM · commercial OK29. Phi 4 mini instruct — microsoft, 3.8B · $0.22/1M API · ~6.0 GB VRAM · commercial OK30. GLM 4.7 Flash — zai-org, 31.2B · $0.23/1M API · ~28 GB VRAM · commercial OK31. Mixtral 8x7B Instruct — Mistral AI, 46.7B · $0.24/1M API · ~34 GB VRAM · commercial OK32. Gemma 3 27B — Google, 27B · $0.27/1M API · ~30 GB VRAM · commercial OK33. Qwen3-8B — Alibaba, 8.2B · $0.29/1M API · ~9.0 GB VRAM · commercial OK34. Qwen3 30B A3B — Qwen, 30.5B · $0.31/1M API · ~22 GB VRAM · commercial OK35. DeepSeek V3.2 — deepseek-ai, 685.4B · $0.32/1M API · ~490 GB VRAM · commercial OK36. Hy3 — tencent, 298.8B · $0.33/1M API · ~204 GB VRAM · commercial OK37. DeepSeek V3.2 Exp — deepseek-ai, 685.4B · $0.34/1M API · ~490 GB VRAM · commercial OK38. Llama 4 Maverick (17B-128E) — Meta, 402B · $0.38/1M API · ~274 GB VRAM · commercial OK39. Hy3 preview — tencent, 298.8B · $0.39/1M API · ~204 GB VRAM · non-commercial40. Llama 3.1 70B Instruct — Meta, 70B · $0.40/1M API · ~53 GB VRAM · commercial OKShowing the top 40 of 69. See all →
More: all "best" lists · Outcome Lab · all models
Open the free LLM cost calculator → · $0 markup. © 2026 Cynosure LLC.