Monthly intelligence for teams running inference in production
Generated 2026-10-22 02:55 UTC from 468 priced models across 232 model families, plus 156 open-weight releases.
7 router/synthetic endpoints with no list price are excluded from the tables below.
No prior snapshot to compare against yet.
| Model | Prompt /1M | Completion /1M | Context |
|---|---|---|---|
| recursive/flint-v1.0:free | free | free | 262,144 |
| minimax/minimax-m3.1-flash:free | free | free | 1,000,000 |
| deepseek/deepseek-v4.1-flash-1017 | $0.3 | $1.2 | 1,048,576 |
| elorian/glimpse | $0.6 | $3 | 262,144 |
| xingchen-agi/xing4.0-29b-a4b:free | free | free | 262,144 |
| reflection/beam-1.0-501b-a23b | $0.3 | $1 | 1,048,576 |
| inception/mercury-3 | $0.1 | $0.35 | 1,000,000 |
| moondream/moondream3-preview | $0.3 | $2.5 | 32,000 |
| moondream/moondream3.1-9b-a2b | $0.3 | $2.5 | 32,000 |
| meta/muse-spark-open | $1.25 | $4.25 | 1,048,576 |
| nvidia/nemotron-3.5-super-vl | $0.15 | $0.5 | 1,048,576 |
| nvidia/nemotron-3.5-super-vl:free | free | free | 1,000,000 |
| unbiased/pareto-26.10 | $0.8 | $3.2 | 1,048,576 |
| ~unbiased/pareto-latest | $0.8 | $3.2 | 1,048,576 |
| unbiased/pareto-26.9 | $2.5 | $7.5 | 262,144 |
| Model | Prompt /1M | Completion /1M | Context |
|---|---|---|---|
| ibm-granite/granite-4.0-h-micro | $0.017 | $0.112 | 131,000 |
| openai/gpt-oss-20b | $0.018 | $0.09 | 131,072 |
| z-ai/glm-5.3-flash | $0.02 | $0.121 | 1,048,576 |
| ~z-ai/glm-flash-latest | $0.02 | $0.121 | 1,048,576 |
| inclusionai/ling-3.0-flash | $0.021 | $0.063 | 262,144 |
| openai/gpt-oss-20b:batch | $0.024 | $0.112 | 131,072 |
| nex-agi/nex-n2.5-mini | $0.025 | $0.1 | 262,144 |
| openai/gpt-5-nano:batch | $0.025 | $0.2 | 400,000 |
| meta-llama/llama-3.2-1b-instruct | $0.027 | $0.201 | 60,000 |
| mistralai/mistral-nemo | $0.029 | $0.03 | 131,072 |
Cost of 1M tokens where 750k are prompt and 250k are completion — the shape of most chat and RAG traffic.
| Model | Blended /1M |
|---|---|
| mistralai/mistral-nemo | $0.029 |
| scaledown/extract | $0.03 |
| scaledown/summarize | $0.03 |
| inclusionai/ling-3.0-flash | $0.032 |
| openai/gpt-oss-20b | $0.036 |
| ibm-granite/granite-4.0-h-micro | $0.041 |
| google/gemma-3-4b-it | $0.043 |
| sao10k/l3-lunaris-8b | $0.043 |
| nex-agi/nex-n2.5-mini | $0.044 |
| z-ai/glm-5.3-flash | $0.045 |
| Model | Context | Prompt /1M |
|---|---|---|
| x-ai/grok-4.20 | 2,000,000 | $1.25 |
| x-ai/grok-4.20-multi-agent | 2,000,000 | $1.25 |
| meta-llama/llama-4-scout | 1,310,720 | $0.1 |
| openai/gpt-5.4 | 1,050,000 | $2.5 |
| openai/gpt-5.4-pro | 1,050,000 | $30 |
| openai/gpt-5.4-pro:batch | 1,050,000 | $15 |
| openai/gpt-5.4:batch | 1,050,000 | $1.25 |
| openai/gpt-5.5 | 1,050,000 | $5 |
| openai/gpt-5.5-pro | 1,050,000 | $30 |
| openai/gpt-5.5-pro:batch | 1,050,000 | $15 |
| Model | Released | Downloads | Likes |
|---|---|---|---|
| RecursiveSI/flint | 2026-10-21 | 0 | 269 |
| InternScience/Sage-A1.5-Q4KM-GGUF | 2026-10-21 | 10 | 32 |
| ATH-MaaS/OvisOCR3 | 2026-10-21 | 7 | 40 |
| KRAFTON/Raon-1-mini-preview | 2026-10-21 | 25 | 107 |
| KRAFTON/Raon-1-mini-preview-gguf | 2026-10-20 | 9 | 26 |
| unsloth/Ornith-1.5-122B-GGUF | 2026-10-20 | 7,758 | 64 |
| unsloth/flint-GGUF | 2026-10-20 | 11 | 73 |
| ornith-ai/Ornith-1.5-122B-GGUF | 2026-10-20 | 7,699 | 37 |
| AtomicChat/DeepSeek-V4.1-Flash-1017-Uncensored-GGUF | 2026-10-19 | 132 | 30 |
| baselabs/mammoth-1 | 2026-10-19 | 21 | 25 |
| briaai/Fibo-2.0-edit-turbo | 2026-10-19 | 310 | 87 |
| gradium/phonon-31m-en | 2026-10-19 | 3,590 | 68 |
| gradium/phonon-40m-multilingual | 2026-10-19 | 349 | 34 |
| gradium/phonon-40m-en | 2026-10-19 | 562 | 54 |
| XHToken/Spark-X2.6-4B-GGUF | 2026-10-19 | 1,359 | 60 |
Prompt and completion prices are per million tokens, taken from the OpenRouter model catalogue. Open-weight entries are Hugging Face releases with real download traction, so the list is filtered to things people actually use rather than every upload.
A price cut on a model you already depend on is the highest-signal event here: it is a direct margin change, and the window to renegotiate or re-route traffic is usually weeks, not months. A new open-weight release matters if it is cheap enough to self-host at your volume — at that point the per-token price is a GPU-hour question instead of an API question.
Prices captured 2026-10-22. Verify against vendor pricing before committing.