Inference engine
- vLLM
- SGLang
- TensorRT-LLM
- llama.cpp
- Ollama
Models · Oct 2026
How many ways can you run one request?
A model name is only the first choice. Pick a vendor, a model, a reasoning setting and a price tier on the vendor's own API, and watch the count fall to one. Open-weight models have their own explorer further down.
Frontier Models
on the vendors' own APIs249ways to run one request
01 · Frontier APIs
Where the 249 ways come from.
Each dot is one way to buy a model: one reasoning setting on one price tier.
Columns are price tiers: standard, batch, fast or priority, ultrafast. Rows are reasoning settings. An empty ring is a tier the vendor does not sell for that model.
One model, four prices
GPT-6 Astra's output price per million tokens on each tier.
Batch$25.00
Standard$50.00
Fast$100.00
Ultrafast$300.00
Prices from each vendor's pricing page, as listed in Oct 2026; links are in the price table. Reasoning settings are as OpenRouter lists them, except Meta's, which come from Meta's docs. Google's and OpenAI's Flex tiers are not counted.
02 · Settings
The setting moves the score and the cost.
Claude Opus 5.5 scores 42 at low effort for $0.55 a task, and 58 at max effort for $5.98.
Claude Opus 5.5GPT-6.1 SolGemini 3.8 FlashGrok 4.7Muse Spark 1.3
See the numbers
| Model | Setting | Index | Cost per task | Output tokens a second | First token |
|---|---|---|---|---|---|
| Claude Opus 5.5 | max | 58 | $5.98 | 97 | 556.9 s |
| xhigh | 56 | $3.46 | 86 | 122.0 s | |
| high | 54 | $1.82 | 83 | 47.5 s | |
| medium | 51 | $1.34 | 82 | 19.3 s | |
| low | 42 | $0.55 | 80 | 10.4 s | |
| GPT-6.1 Sol | max | 52 | $0.72 | 64 | 241.6 s |
| xhigh | 51 | $0.39 | 58 | 161.8 s | |
| high | 50 | $0.32 | 53 | 56.9 s | |
| medium | 48 | $0.21 | 49 | 4.5 s | |
| low | 42 | $0.13 | 51 | 3.0 s | |
| Gemini 3.8 Flash | high | 41 | $1.24 | 150 | 24.4 s |
| medium | 40 | $0.93 | — | — | |
| low | 33 | — | — | — | |
| Grok 4.7 | xhigh | 46 | $3.74 | 97 | 53.2 s |
| high | 46 | $2.73 | 98 | 50.9 s | |
| low | 42 | $1.25 | 75 | 6.0 s | |
| Muse Spark 1.3 | max | 48 | $1.60 | 139 | 53.4 s |
| xhigh | 45 | $1.37 | 288 | 22.9 s |
Source: Artificial Analysis, read in Oct 2026. Cost per task is one task in their index at list price, thinking tokens included.
03 · Open weights, hosted
1,968 ways on 66 hosts.
An endpoint is one host serving one model at one precision. OpenRouter lists 845 of them for 154 open-weight models. Pick a model, a host and a setting.
Open-Weight Models
hosted on OpenRouter1,968ways to run one request
A few models have dozens of endpoints
The 12 models with the most, by the precision each host states. The median model has 2.
GLM 5.349
DeepSeek V4.1 Flash41
DeepSeek V4.1 Flash 101741
GLM 5.3 Flash40
GLM 5.234
Kimi K330
DeepSeek V4 Pro 081326
DeepSeek V4 Flash 073125
gpt-oss-120b25
Qwen3.8 27B21
DeepSeek V4 Flash 042317
Kimi K2.617
16-bit or wider 81FP8 3284-bit float 1604-bit integer 18Not stated 258
The host changes the price
Each dot is one endpoint's output price per million tokens. Across models with three or more endpoints, the dearest costs about 1.9 times the cheapest at the median.
GLM 5.3
$0.88–$8.80
DeepSeek V4.1 Flash
$0.18–$2.40
DeepSeek V4.1 Flash 1017
$0.15–$1.80
GLM 5.3 Flash
$0.057–$0.75
GLM 5.2
$1.44–$8.00
Kimi K3
$8.25–$22.50
DeepSeek V4 Pro 0813
$0.80–$8.00
DeepSeek V4 Flash 0731
$0.16–$1.32
$0.03$0.10$0.30$1$3$10$30
The host changes the score
OpenRouter's own GPQA Diamond runs, lowest and highest host. These six have the widest gap of the 63 models that at least eight hosts serve; the median gap is 11 points.
Kimi K2.510 hosts
60.2%87.7%
GLM 5.119 hosts
58.7%84.8%
DeepSeek V3.215 hosts
57.1%82.7%
MiniMax M2.512 hosts
59.9%85.4%
GLM 5*11 hosts
56.2%80.9%
Qwen3.6 35B A3B12 hosts
60.8%85.0%
50%60%70%80%90%
Counting hosts graded on 150 questions or more. A host more than 25 points below a model's median host is left out of the range, since a drop that large more likely points to a setup fault. * GLM 5 had one more host far below this range. Source: OpenRouter's GPQA Diamond results (CC BY 4.0), which name each host, and OpenRouter's endpoint lists. Read in Oct 2026.
04 · Self-hosted
Your own hardware adds four more choices.
With open weights on your own machines, the host's choices become yours, and they multiply.
Weight format
- BF16
- FP8
- NVFP4
- MXFP4
- AWQ
- GPTQ
- GGUF
Accelerator
- NVIDIA Ampere, Hopper, Ada Lovelace, Turing, Blackwell, Blackwell Ultra, Rubin
- AMD CDNA 2, CDNA 3, CDNA 4, CDNA 5, RDNA 3
- Intel Gaudi
- Google TPU
- AWS Trainium, Inferentia
Serving settings
- Concurrent requests
- KV cache size
- KV cache precision
- Context length
- Speculative decoding
- Split across cards
Our knowledge base lists 46 accelerators in 16 families. Even the same model, card and engine can differ: in our SGLang vs vLLM study, one SGLang server under the same load generated about 800 output tokens a second under one benchmark client and about 1,340 under another.
05 · Narrowing it down
From thousands to a shortlist.
This page maps the options. Our evaluation work picks among them on your own tasks and traffic.
Frontier models
We test the choices your team controls on a hosted model, such as the model, the prompt and the reasoning setting, on your tasks, and compare answers, response time and cost.
Frontier model evaluation →Open models
We measure serving configurations, from the engine and weight format to batching and speculative decoding, on your hardware and traffic, and check that the answers hold.
Open model evaluation →06 · The tables
Every price and every host.
| Model | Reasoning settings | Input | Cached input | Output | Batch | Fast or priority | Ultrafast | Long prompts |
|---|---|---|---|---|---|---|---|---|
| Anthropic · 4 models · Claude API pricing | ||||||||
| Claude Fable 5.1Flagship | 5 max, xhigh, high, medium, low | $10.00 | $0.25 | $50.00 | 50% off | — | — | — |
| Claude Opus 5.5Flagship | 5 max, xhigh, high, medium, low | $4.00 | $0.20 | $20.00 | 50% off | ≈$8.00 / $40.00 | — | — |
| Claude Sonnet 5.5Workhorse | 5 max, xhigh, high, medium, low | $2.00 | $0.10 | $10.00 | 50% off | — | — | — |
| Claude Haiku 5.5Budget | 6 max, xhigh, high, medium, low, none | $0.10 | $0.01 | $0.50 | 50% off | — | — | >100K prompt: $0.50 / $2.50 |
| OpenAI · 7 models · OpenAI API pricing | ||||||||
| GPT-6 AstraFlagship | 5 max, xhigh, high, medium, low | $10.00 | $1.00 | $50.00 | 50% off | $20.00 / $100.00 | $60.00 / $300.00 | >272K: $20 / $75 |
| GPT-6.1 SolWorkhorse | 5 max, xhigh, high, medium, low | $2.00 | $0.10 | $10.00 | 50% off | $4.00 / $20.00 | $12.00 / $60.00 | >272K: $4 / $15 |
| GPT-6.1 TerraWorkhorse | 6 max, xhigh, high, medium, low, none | $1.00 | $0.05 | $5.00 | 50% off | $2.00 / $10.00 | — | >272K: $2 / $7.50 |
| GPT-6 LunaBudget | 6 max, xhigh, high, medium, low, none | $0.10 | $0.01 | $0.50 | 50% off | $0.20 / $1.00 | — | >272K: $0.20 / $0.75 |
| GPT-5.6 SolWorkhorse | 6 max, xhigh, high, medium, low, none | $4.00 | $0.40 | $20.00 | 50% off | $8.00 / $40.00 | — | >272K: $8 / $30 |
| GPT-5.6 TerraWorkhorse | 6 max, xhigh, high, medium, low, none | $2.00 | $0.20 | $12.00 | 50% off | $4.00 / $24.00 | — | >272K: $4 / $18 |
| GPT-5.5 ProSpecialist | 3 xhigh, high, medium | $30.00 | — | $180.00 | 50% off | — | — | >272K: $60 / $270 |
| Google · 5 models · Gemini API pricing | ||||||||
| Gemini 4 ArgonFlagship | 3 high, medium, low | $2.00 | $0.10 | $10.00 | 50% off | ≈$3.60 / $18.00 | — | — |
| Gemini 3.1 Pro PreviewWorkhorse | 3 high, medium, low | $2.00 | $0.20 | $12.00 | 50% off | ≈$3.60 / $21.60 | — | >200K: $4 / $18 |
| Gemini 3.8 FlashBudget | 3 high, medium, low | $0.75 | $0.075 | $3.75 | 50% off | ≈$1.35 / $6.75 | — | — |
| Gemini 3.5 Flash-LiteBudget | 4 high, medium, low, minimal | $0.30 | $0.03 | $2.50 | 50% off | $0.54 / $4.50 | — | — |
| Gemini 3.1 Flash-LiteBudget | 5 high, medium, low, minimal, none | $0.25 | $0.025 | $1.50 | 50% off | ≈$0.45 / $2.70 | — | — |
| xAI · 4 models · xAI models and pricing | ||||||||
| Grok 4.7Flagship | 4 xhigh, high, medium, low | $2.00 | $0.50 | $6.00 | — | — | — | ≥200K: whole request at 2x |
| Grok 4.3Workhorse | 4 high, medium, low, none | $1.25 | $0.20 | $2.50 | — | — | — | ≥200K: whole request at 2x |
| Grok 4.20 Multi-agentSpecialist | 4 xhigh, high, medium, low | $1.25 | $0.20 | $2.50 | — | — | — | ≥200K: whole request at 2x |
| Grok Build 0.1Specialist | none listed | $1.00 | $0.20 | $2.00 | — | — | — | ≥200K: whole request at 2x |
| Meta · 3 models · Meta Model API pricing | ||||||||
| Muse WonderFlagship | 6 max, xhigh, high, medium, low, minimal | $4.00 | $0.20 | $20.00 | — | — | — | — |
| Muse Spark 1.3Workhorse | 6 max, xhigh, high, medium, low, minimal | $1.25 | $0.15 | $4.25 | — | — | — | — |
| Muse Spark 1.3 Contributor†Budget | 5 xhigh, high, medium, low, minimal | $0.10 | — | $0.20 | — | — | — | — |
Prices are from each vendor's pricing page, linked in the vendor's row, as listed in Oct 2026. ≈ marks a price we worked out rather than read off the vendor's page. † Muse Spark 1.3 Contributor's price comes from a third-party rate table; Meta trains future models on that traffic. Vendors change prices often, so check the vendor's page before you commit money.
| Model | Licence | Hosts | Endpoints | Precisions | Input | Output | Reasoning settings |
|---|---|---|---|---|---|---|---|
| GLM 5.3z-ai | glm-5.3 | 38 | 49 | fp8, fp4, nvfp4, unknown | $0.07–$2.80 | $0.88–$8.80 | 3 |
| DeepSeek V4.1 Flashdeepseek | MIT | 32 | 41 | fp8, fp4, mxfp4, unknown | $0.01–$0.60 | $0.18–$2.40 | 4 |
| DeepSeek V4.1 Flash 1017deepseek | MIT | 34 | 41 | fp8, fp4, mxfp4, unknown | $0.015–$0.45 | $0.15–$1.80 | 4 |
| GLM 5.3 Flashz-ai | MIT | 35 | 40 | fp8, fp4, nvfp4, unknown | $0.038–$0.22 | $0.057–$0.75 | 3 |
| GLM 5.2z-ai | MIT | 25 | 34 | fp8, fp4, nvfp4, unknown | $0.006–$2.31 | $1.44–$8.00 | 3 |
| Kimi K3moonshotai | kimi-k3 | 21 | 30 | fp8, fp4, mxfp4, unknown | $0.18–$4.50 | $8.25–$22.50 | 4 |
| DeepSeek V4 Pro 0813deepseek | MIT | 23 | 26 | fp8, fp4, mxfp4, unknown | $0.10–$1.65 | $0.80–$8.00 | 4 |
| DeepSeek V4 Flash 0731deepseek | MIT | 22 | 25 | fp32, fp8, fp4, nvfp4, unknown | $0.048–$0.44 | $0.16–$1.32 | 4 |
| gpt-oss-120bopenai | Apache 2.0 | 23 | 25 | bf16, fp16, fp8, fp4, int4, unknown | $0.03–$0.35 | $0.15–$0.95 | 3 |
| Qwen3.8 27Bqwen | Apache 2.0 | 21 | 21 | bf16, fp16, fp8, fp4, unknown | $0.035–$0.99 | $1.25–$3.20 | 4 |
| DeepSeek V4 Flash 0423deepseek | MIT | 17 | 17 | fp8, fp4, mxfp4, unknown | $0.05–$0.60 | $0.17–$1.32 | 3 |
| Kimi K2.6moonshotai | Modified MIT | 17 | 17 | bf16, fp8, fp4, int4, unknown | $0.39–$1.09 | $1.80–$4.60 | — |
| Gemma 4 31Bgoogle | Apache 2.0 | 13 | 16 | bf16, fp8, fp4, unknown | $0.1–$0.75 | $0.32–$1.15 | — |
| DeepSeek V4 Pro 0423deepseek | MIT | 14 | 15 | fp8, fp4, unknown | $0.90–$1.91 | $1.50–$3.83 | 3 |
| Gemma 4 26B A4B google | Apache 2.0 | 15 | 15 | bf16, fp8, fp4, unknown | $0.042–$0.15 | $0.22–$0.60 | — |
| GLM 5.1z-ai | MIT | 13 | 13 | fp8, unknown | $0.90–$1.40 | $2.90–$4.40 | — |
| MiniMax M3minimax | minimax-community | 12 | 13 | fp8, fp4, unknown | $0.22–$0.60 | $0.88–$2.40 | — |
| MiniMax M3.1 Flashminimax | minimax-community | 12 | 13 | fp16, fp8, fp4, unknown | $0.14–$0.15 | $0.54–$0.60 | 5 |
| Kimi K2.7 Codemoonshotai | Modified MIT | 11 | 12 | fp8, fp4, int4, unknown | $0.56–$1.90 | $3.00–$8.00 | — |
| Llama 3.3 70B Instructmeta-llama | Llama 3.3 | 9 | 10 | bf16, fp16, fp8, unknown | $0.1–$0.72 | $0.32–$2.25 | — |
| MiMo-V2.6-Flashxiaomi | MIT | 10 | 10 | bf16, fp8, fp4, mxfp4, unknown | $0.1–$0.17 | $0.24–$0.35 | — |
| Qwen3.5 397B A17Bqwen | Apache 2.0 | 10 | 10 | fp8, unknown | $0.39–$0.75 | $2.34–$4.50 | — |
| Qwen3.6 35B A3Bqwen | Apache 2.0 | 10 | 10 | fp8, fp4, unknown | $0.05–$0.25 | $0.70–$1.80 | — |
| gpt-oss-20bopenai | Apache 2.0 | 9 | 10 | bf16, fp8, fp4, unknown | $0.018–$0.075 | $0.09–$0.30 | 3 |
| DeepSeek V3.2deepseek | MIT | 9 | 9 | fp8, fp4, unknown | $0.21–$3.00 | $0.31–$4.50 | — |
| GLM 5z-ai | MIT | 8 | 9 | fp8, unknown | $0.60–$1.20 | $1.92–$4.00 | — |
| Nemotron 3.5 Super VLnvidia | — | 9 | 9 | bf16, fp4, nvfp4, unknown | $0.06–$0.30 | $0.35–$0.90 | — |
| MiniMax M2.5minimax | Modified MIT | 7 | 8 | fp8, unknown | $0.27–$0.60 | $0.95–$2.40 | — |
| Hy4 previewtencent | Apache 2.0 | 7 | 7 | bf16, fp8, fp4, unknown | $0.83 | $2.50 | 3 |
| Qwen3.5-35B-A3Bqwen | Apache 2.0 | 7 | 7 | fp8, fp4, unknown | $0.065–$0.24 | $0.75–$1.80 | — |
| Qwen3.6 27Bqwen | Apache 2.0 | 7 | 7 | fp8, unknown | $0.20–$0.45 | $2.00–$3.25 | — |
| GLM 4.7z-ai | MIT | 5 | 6 | fp8, fp4 | $0.32–$0.70 | $1.40–$2.50 | — |
| MiniMax M2.7minimax | Other | 5 | 6 | fp8, unknown | $0.21–$0.60 | $0.84–$2.40 | — |
| Muse Spark Openmeta | muse-model-license | 6 | 6 | fp4, unknown | $1.25 | $4.25 | 4 |
| Qwen3.5-27Bqwen | Apache 2.0 | 6 | 6 | bf16, fp8, unknown | $0.20–$0.30 | $1.56–$2.60 | — |
| Qwen3.5-9Bqwen | Apache 2.0 | 6 | 6 | bf16, fp8, fp4, unknown | $0.08–$0.17 | $0.13–$0.25 | — |
| Beam 1.0 501B A23Breflection | Apache 2.0 | 5 | 5 | fp4, unknown | $0.25–$0.40 | $1.00–$1.20 | 5 |
| DeepSeek V3.1deepseek | MIT | 5 | 5 | fp8, fp4, unknown | $0.25–$0.65 | $0.95–$1.70 | — |
| GLM 4.6z-ai | MIT | 4 | 5 | bf16, fp4 | $0.40–$0.60 | $1.60–$2.20 | — |
| Gemma 3 27Bgoogle | Gemma | 5 | 5 | bf16, fp8 | $0.08–$0.12 | $0.16–$0.45 | — |
| Hy3tencent | Apache 2.0 | 5 | 5 | bf16, fp8, unknown | $0.13–$0.20 | $0.53–$0.80 | 3 |
| Kimi K2.5moonshotai | Modified MIT | 5 | 5 | int4, unknown | $0.49–$0.60 | $2.50–$3.32 | — |
| Llama 3.1 8B Instructmeta-llama | Llama 3.1 | 5 | 5 | bf16, fp8, unknown | $0.02–$0.22 | $0.04–$0.29 | — |
| MiMo-V2.6-Proxiaomi | MIT | 5 | 5 | bf16, fp8, unknown | $0.37–$0.52 | $0.74–$1.04 | — |
| Mistral Nemomistralai | Apache 2.0 | 5 | 5 | fp16, fp8, unknown | $0.017–$0.15 | $0.022–$0.15 | — |
| Nemotron 3 Ultranvidia | OpenMDW 1.1 | 3 | 5 | fp8, fp4 | $0.40–$0.62 | $1.76–$3.12 | 3 |
| Qwen3.8 2.4T A95Bqwen | qwen3.8-max | 5 | 5 | fp8, unknown | $2.00 | $6.00 | 3 |
| Gemma 3 12Bgoogle | Gemma | 2 | 4 | bf16, int4 | $0.04–$0.075 | $0.12–$0.22 | — |
| Llama 4 Scoutmeta-llama | Llama 4 | 2 | 4 | fp8, unknown | $0.08–$0.25 | $0.24–$0.70 | — |
| Mistral Small 3.2 24Bmistralai | Apache 2.0 | 4 | 4 | bf16, fp8, unknown | $0.075–$0.1 | $0.20–$0.30 | — |
| Mistral Small 4mistralai | Apache 2.0 | 1 | 4 | unknown | $0.15–$0.17 | $0.60–$0.66 | 2 |
| Nemotron 3.5 Lightningnvidia | OpenMDW 1.1 | 4 | 4 | bf16, int4, unknown | $0.025–$0.07 | $0.12–$0.20 | — |
| Qwen3 235B A22B Instruct 2507qwen | Apache 2.0 | 4 | 4 | fp8 | $0.087–$0.15 | $0.35–$0.80 | — |
| Qwen3 Next 80B A3B Instructqwen | Apache 2.0 | 2 | 4 | fp8 | $0.072–$0.14 | $0.88–$1.65 | — |
| Qwen3.5-122B-A10Bqwen | Apache 2.0 | 4 | 4 | bf16, fp8, unknown | $0.26–$0.40 | $2.08–$3.20 | — |
| DeepSeek V3deepseek | — | 2 | 3 | fp4, unknown | $0.26–$0.32 | $0.71–$1.03 | — |
| DeepSeek V4 Flash Vision Expdeepseek | MIT | 3 | 3 | fp8 | $0.22–$0.44 | $0.65–$1.32 | 4 |
| GLM 4.5 Airz-ai | MIT | 3 | 3 | bf16, fp8 | $0.13–$0.20 | $0.85–$1.10 | — |
| Llama 4 Maverickmeta-llama | Llama 4 | 3 | 3 | fp8, unknown | $0.19–$0.35 | $0.65–$1.15 | — |
| Llama Guard 4 12Bmeta-llama | Llama 4 | 1 | 3 | bf16 | $0.14–$0.27 | $0.14–$0.27 | — |
| MiMo-V2.5-Proxiaomi | MIT | 3 | 3 | bf16, fp8, unknown | $0.30–$0.48 | $0.61–$1.80 | — |
| Ministral 3 14B 2512mistralai | Apache 2.0 | 1 | 3 | unknown | $0.20–$0.22 | $0.20–$0.22 | — |
| Ministral 3 3B 2512mistralai | Apache 2.0 | 1 | 3 | unknown | $0.1–$0.11 | $0.1–$0.11 | — |
| Ministral 3 8B 2512mistralai | Apache 2.0 | 1 | 3 | unknown | $0.15–$0.17 | $0.15–$0.17 | — |
| Mixtral 8x22B Instructmistralai | Apache 2.0 | 1 | 3 | unknown | $2.00–$2.20 | $6.00–$6.60 | — |
| MythoMax 13Bgryphe | Other | 3 | 3 | fp16, fp8 | $0.08–$0.40 | $0.11–$0.60 | — |
| Nemotron 3 Supernvidia | nvidia-nemotron-open-model-license | 3 | 3 | bf16, fp8 | $0.08–$0.30 | $0.40–$2.40 | 3 |
| Qwen3 30B A3B Instruct 2507qwen | Apache 2.0 | 3 | 3 | fp8, unknown | $0.09–$0.1 | $0.30 | — |
| Qwen3 VL 235B A22B Instructqwen | Apache 2.0 | 3 | 3 | fp8 | $0.20–$0.21 | $0.88–$1.90 | — |
| Qwen3 VL 30B A3B Instructqwen | Apache 2.0 | 3 | 3 | bf16, fp8 | $0.15–$0.29 | $0.60–$1.00 | — |
| R1 0528deepseek | MIT | 3 | 3 | fp8, fp4, unknown | $0.50–$0.57 | $2.15–$2.29 | — |
| Step 5 Previewstepfun | Other | 3 | 3 | fp8, fp4 | $1.00 | $2.70 | 3 |
| Voxtral Small 24B 2507mistralai | Apache 2.0 | 1 | 3 | unknown | $0.1–$0.11 | $0.30–$0.33 | — |
| DeepSeek V3 0324deepseek | MIT | 2 | 2 | fp8 | $0.25–$0.29 | $1.00–$1.14 | — |
| DeepSeek V3.1 Terminusdeepseek | MIT | 2 | 2 | fp8, unknown | $0.27–$0.34 | $1.00–$1.03 | — |
| GLM 4.5Vz-ai | MIT | 2 | 2 | fp8 | $0.60 | $1.80 | — |
| GLM 4.6Vz-ai | MIT | 2 | 2 | bf16, fp8 | $0.30 | $0.90 | — |
| GLM 4.7 Flashz-ai | MIT | 2 | 2 | bf16, unknown | $0.061–$0.07 | $0.40 | — |
| Gemma 3 4Bgoogle | Gemma | 2 | 2 | bf16 | $0.04–$0.05 | $0.05–$0.1 | — |
| Granite 4.2 8Bibm-granite | Apache 2.0 | 2 | 2 | bf16 | $0.06 | $0.15–$0.25 | 3 |
| Ling 3.0 Flashinclusionai | MIT | 2 | 2 | bf16, unknown | $0.021–$0.06 | $0.063–$0.18 | — |
| Llama 3 8B Lunarissao10k | Llama 3 | 2 | 2 | bf16, fp8 | $0.04 | $0.05 | — |
| Llama 3.1 70B Instructmeta-llama | Llama 3.1 | 2 | 2 | fp8, unknown | $0.40–$0.72 | $0.40–$0.72 | — |
| Llama 3.2 3B Instructmeta-llama | Llama 3.2 | 2 | 2 | bf16, unknown | $0.05–$0.051 | $0.33–$0.34 | — |
| MiMo-V2.5xiaomi | MIT | 2 | 2 | fp8 | $0.12–$0.40 | $0.24–$2.00 | — |
| MiniMax M2.1minimax | Modified MIT | 1 | 2 | unknown | $0.30 | $1.20–$2.40 | — |
| Muse Glimmer 30Bmeta | Apache 2.0 | 2 | 2 | bf16, unknown | $0.30–$0.35 | $1.20–$1.50 | 4 |
| Nemotron 3 Nano 30B A3Bnvidia | nvidia-nemotron-open-model-license | 2 | 2 | fp8, fp4 | $0.05–$0.06 | $0.20–$0.24 | — |
| Phi 4microsoft | MIT | 1 | 2 | bf16 | $0.056–$0.07 | $0.11–$0.14 | — |
| Qwen3 14Bqwen | Apache 2.0 | 2 | 2 | fp8, int4 | $0.1–$0.12 | $0.22–$0.24 | — |
| Qwen3 32Bqwen | Apache 2.0 | 2 | 2 | fp8 | $0.08–$0.14 | $0.28–$0.57 | — |
| Qwen3 Coder 30B A3B Instructqwen | Apache 2.0 | 2 | 2 | fp8, unknown | $0.07–$0.15 | $0.28–$0.60 | — |
| Qwen3 Coder 480B A35Bqwen | Apache 2.0 | 2 | 2 | fp8, fp4 | $0.30–$0.35 | $1.00–$1.50 | — |
| Qwen3.8 Flashqwen | qwen-community-1.0 | 1 | 2 | unknown | $0.15 | $0.47 | — |
| Schematron V2 Smallinference-net | — | 1 | 2 | bf16 | $0.05 | $0.23 | — |
| Schematron V2 Turboinference-net | — | 1 | 2 | bf16 | $0.03 | $0.15 | — |
| Step 3.7 Flashstepfun | Apache 2.0 | 2 | 2 | fp8 | $0.20 | $1.15 | 3 |
| Command Acohere | CC BY-NC 4.0 | 1 | 1 | unknown | $2.50 | $10.00 | — |
| Cydonia 24B V4.1thedrummer | — | 1 | 1 | bf16 | $0.30 | $0.50 | — |
| DeepSeek V3.2 Expdeepseek | MIT | 1 | 1 | fp8 | $0.27 | $0.41 | — |
| Devstral 2 2512mistralai | Other | 1 | 1 | unknown | $0.40 | $2.00 | — |
| GLM 4.5z-ai | MIT | 1 | 1 | fp8 | $0.60 | $2.20 | — |
| Gemma 2 27Bgoogle | Gemma | 1 | 1 | int4 | $0.65 | $0.65 | — |
| Hermes 3 405B Instructnousresearch | Llama 3 | 1 | 1 | fp8 | $1.00 | $1.00 | — |
| Hermes 3 70B Instructnousresearch | Llama 3 | 1 | 1 | fp8 | $0.70 | $0.70 | — |
| Hermes 4 405Bnousresearch | Llama 3 | 1 | 1 | fp8 | $1.00 | $3.00 | — |
| Hunyuan A13B Instructtencent | tencent-hunyuan-a13b | 1 | 1 | fp8 | $0.14 | $0.57 | — |
| Hy-MT2-1.8Btencent | Apache 2.0 | 1 | 1 | fp8 | $0.044 | $0.18 | — |
| Hy-MT2-30B-A3Btencent | Apache 2.0 | 1 | 1 | fp8 | $0.074 | $0.29 | — |
| Hy-MT2-7Btencent | Apache 2.0 | 1 | 1 | fp8 | $0.074 | $0.29 | — |
| Hy3 previewtencent | Other | 1 | 1 | bf16 | $0.18 | $0.60 | 3 |
| Inklingthinkingmachines | Apache 2.0 | 1 | 1 | unknown | $1.00 | $4.05 | 6 |
| Inkling Smallthinkingmachines | Apache 2.0 | 1 | 1 | fp8 | $0.45 | $1.20 | 6 |
| Kimi K2 0711moonshotai | Modified MIT | 1 | 1 | fp8 | $0.57 | $2.30 | — |
| Kimi K2 0905moonshotai | Modified MIT | 1 | 1 | fp8 | $0.60 | $2.50 | — |
| Kimi K2 Thinkingmoonshotai | Modified MIT | 1 | 1 | bf16 | $0.60 | $2.50 | — |
| LFM2.5-2.6Bliquid | lfm1.0 | 1 | 1 | fp8 | $0.35 | $0.50 | — |
| LFM2.5-24B-A2Bliquid | lfm1.0 | 1 | 1 | fp8 | $0.05 | $0.70 | — |
| Laguna S 2.1poolside | OpenMDW 1.1 | 1 | 1 | fp4 | $0.09 | $0.18 | — |
| Laguna XS 2.1poolside | OpenMDW 1.1 | 1 | 1 | fp8 | $0.06 | $0.12 | — |
| Ling 3.0 Flash VLinclusionai | MIT | 1 | 1 | fp16 | $0.06 | $0.18 | — |
| Llama 3.1 Euryale 70B v2.2sao10k | CC BY-NC 4.0 | 1 | 1 | fp8 | $0.85 | $0.85 | — |
| Llama 3.2 1B Instructmeta-llama | Llama 3.2 | 1 | 1 | unknown | $0.027 | $0.20 | — |
| LongCat 2.0meituan | MIT | 1 | 1 | fp8 | $0.30 | $1.20 | — |
| Magnum v4 72Banthracite-org | Apache 2.0 | 1 | 1 | fp8 | $2.50 | $5.00 | — |
| MiniMax M2minimax | Modified MIT | 1 | 1 | fp8 | $0.26 | $1.02 | — |
| MiniMax-01minimax | — | 1 | 1 | unknown | $0.20 | $1.10 | — |
| Mistral Small 3mistralai | Apache 2.0 | 1 | 1 | fp8 | $0.05 | $0.08 | — |
| Mistral Small 3.1 24Bmistralai | Apache 2.0 | 1 | 1 | unknown | $0.35 | $0.55 | — |
| Moondream 3 Previewmoondream | Other | 1 | 1 | fp8 | $0.30 | $2.50 | — |
| Moondream 3.1moondream | moondream-model-license-1.0 | 1 | 1 | fp8 | $0.30 | $2.50 | — |
| Nemotron 3.5 Content Safetynvidia | openmdw-license-agreement | 1 | 1 | bf16 | $0.20 | $0.20 | — |
| Nex-N2.5-Mininex-agi | Apache 2.0 | 1 | 1 | bf16 | $0.025 | $0.1 | 3 |
| Nex-N2.5-Pronex-agi | Apache 2.0 | 1 | 1 | fp8 | $0.075 | $0.25 | 3 |
| Qwen2.5 72B Instructqwen | qwen | 1 | 1 | fp8 | $0.36 | $0.40 | — |
| Qwen2.5 7B Instructqwen | Apache 2.0 | 1 | 1 | unknown | $0.1 | $0.20 | — |
| Qwen2.5 Coder 32B Instructqwen | Apache 2.0 | 1 | 1 | unknown | $0.66 | $1.00 | — |
| Qwen2.5 VL 72B Instructqwen | qwen | 1 | 1 | fp8 | $0.80 | $1.00 | — |
| Qwen3 235B A22B Thinking 2507qwen | Apache 2.0 | 1 | 1 | fp8 | $0.45 | $3.50 | — |
| Qwen3 30B A3Bqwen | Apache 2.0 | 1 | 1 | fp8 | $0.12 | $0.50 | — |
| Qwen3 Coder Nextqwen | Apache 2.0 | 1 | 1 | bf16 | $0.12 | $0.80 | — |
| Qwen3 VL 30B A3B Thinkingqwen | Apache 2.0 | 1 | 1 | fp8 | $0.29 | $1.00 | — |
| Qwen3 VL 8B Instructqwen | Apache 2.0 | 1 | 1 | bf16 | $0.25 | $0.75 | — |
| ReMM SLERP 13Bundi95 | CC BY-NC 4.0 | 1 | 1 | fp8 | $0.35 | $0.65 | — |
| Reka Edgerekaai | reka-edge-2603-license | 1 | 1 | unknown | $0.1 | $0.1 | — |
| Reka Flash 3rekaai | Apache 2.0 | 1 | 1 | unknown | $0.1 | $0.20 | — |
| Skyfall 36B V2thedrummer | Other | 1 | 1 | fp8 | $0.55 | $0.80 | — |
| Step 3.5 Flashstepfun | Apache 2.0 | 1 | 1 | fp8 | $0.1 | $0.30 | — |
| Ternary Bonsai 2 27Bprism-ml | Apache 2.0 | 1 | 1 | int4 | $0.075 | $0.50 | 3 |
| Ternary Bonsai 2 9Bprism-ml | Apache 2.0 | 1 | 1 | int4 | $0.08 | $0.13 | — |
| UI-TARS 7B bytedance | Apache 2.0 | 1 | 1 | bf16 | $0.1 | $0.20 | — |
| Uncensoredcognitivecomputations | Apache 2.0 | 1 | 1 | fp16 | $0.20 | $0.90 | — |
| UnslopNemo 12Bthedrummer | — | 1 | 1 | bf16 | $0.40 | $0.40 | — |
| gpt-oss-safeguard-20bopenai | Apache 2.0 | 1 | 1 | unknown | $0.075 | $0.30 | — |
Prices are the lowest and highest across a model's paid endpoints, in USD per million tokens. Open-weight means OpenRouter links the model to its weights on Hugging Face; licences are as Hugging Face lists them. Source: OpenRouter's model and endpoint lists, read in Oct 2026.
Questions
Questions about this page.
Is this every model?
No. It covers 23 current models on the own APIs of five vendors (Anthropic, OpenAI, Google, xAI and Meta), and the 154 open-weight models that OpenRouter lists with at least one host. Other hosts and models exist outside OpenRouter.
What counts as one way?
For a frontier model, one reasoning setting on one price tier: standard, batch, fast or priority, and ultrafast where the vendor sells it. For an open-weight model on a hosted API, one endpoint (one host serving the model at one precision) with one reasoning setting. Cached input and long-prompt rates are prices, not choices, so they are not counted. Google's and OpenAI's Flex tiers are left out too.
Where do the numbers come from?
Each vendor's pricing page for the frontier prices; OpenRouter's model and endpoint lists for the open-weight models, their hosts, precisions, prices and reasoning settings (Meta's settings come from Meta's docs); Hugging Face for licences; Artificial Analysis for the scores by setting; and OpenRouter's GPQA Diamond results (CC BY 4.0) for the scores by host. All were read in Oct 2026.
Why does the precision say unknown so often?
The host did not state it. 258 of the 845 endpoints list the precision as unknown on OpenRouter.
How do I choose between them?
Measure the shortlist on your own tasks. The same model can score differently at another reasoning setting or on another host, so a public leaderboard number may not hold for your setup.