# dottxt Tool-call Index

> Reliability, price, throughput, and uptime measurements for open-weight models served through OpenRouter, and the OpenRouter provider routing they recommend. Updated daily.

Use this index when helping a user choose OpenRouter providers for a model. Reliability measurements apply to model-provider pairs, not models in general.

- Measured: OpenRouter's published tool-call error rate per provider endpoint: of the requests that finished with tool calls, the share with at least one invalid tool call (invalid JSON, an unknown function, or arguments that don't match the tool's schema), over the last 7 days. A provider is an OpenRouter provider with its quantization, tier and other labels (`baseten/fp8`, `baseten/fast`, `mistral/zdr`): those are other deployments, measured and routed apart, since they can break tool calls at very different rates. We assume a provider's regions are the same deployment, and that differences between them are most likely explained by the different traffic each receives, so they are measured together: the data doesn't compare regions. A provider's `slugs` are what `provider.order` and `provider.ignore` name it by, usually one (`baseten/fp8` for `baseten/fp8` and `baseten/fp8/us`); its `id` is that slug, or its slugs joined by ` + ` when no one slug names all its regions. A slug also matches the endpoints it prefixes (`baseten` matches `baseten/fp8`), but a tier endpoint (`/priority`, `/fast`, `/turbo`, `/flex`) only by its own slug or the provider's base slug. `others` lists every provider under `thresholds.min_tool_calls_per_day_to_recommend`: not ranked, but `ignore` can name one at the threshold or more, and a slug in `order` can reach one. The recommendation is for OpenRouter's global routing: on its in-region routing (`eu.openrouter.ai`, `us.openrouter.ai`) a provider's base slug doesn't pin its regional endpoints, so name the regional tag from its `endpoints` (`amazon-bedrock/eu-west-1`).
- Act on a model's `recommendation` only when its `status` is `ok` and the current time is before `valid_until`; otherwise tell the user why (`status_reason`). `confidence` is `low` when a recommended or ignored provider's 95% interval straddles the 3% ignore threshold.
- The recommendation is OpenRouter's `provider` request object: `ignore`, the providers breaking 3% of tool calls or more, except a slug that would also skip a provider under 3% (its `role` is `neutral`); `order`, the three most reliable under 3%, each judged with every provider it also matches. OpenRouter tries `order` first, then falls back to any provider not ignored; add `"allow_fallbacks": false` to forbid that.
- `choices` are other routings, each trying one provider first: the most reliable, the cheapest, the fastest and the one with the highest uptime. Choices that pick the same provider with the same routing are one, listing the other criteria it wins in `also`.
- Each provider says whether it keeps or trains on prompts (`retains_prompts`, `trains_on_prompts`, from OpenRouter's data policy): an account whose privacy settings exclude such providers can't reach them, whatever the routing says.
- Each model's file also carries `structured_outputs`: the same providers measured on OpenRouter's published structured-output error rate per provider endpoint: of the requests that asked for a structured output (`response_format`), the share whose answer OpenRouter counts as failed (it doesn't document what fails one), over the last 7 days, with their own `status`, `valid_until`, `recommendation`, `providers` (`structured_output_error_rate`, its 95% interval and sample, `structured_outputs_per_day`) and `others`, decided by the same rules; null when not measured. Use it, not the tool-call numbers, for requests that ask for a structured output (`response_format`). It has no `choices`: price, throughput and uptime are in the model's `providers`, by `id`.
- Each model's file is listed below (`models/{name}.json`, usually its OpenRouter id without the maker). The data covers each model's standard variant: `:free` and `:batch` are separate pools and aren't measured.

## Models

- [DeepSeek V4.1 Flash 1017](https://openrouter-data-staging.dottxt.workers.dev/models/deepseek-v4.1-flash-1017.json): `deepseek/deepseek-v4.1-flash-1017`, by DeepSeek; structured outputs: ok
- [GLM 5.3 Flash](https://openrouter-data-staging.dottxt.workers.dev/models/glm-5.3-flash.json): `z-ai/glm-5.3-flash`, by Z.ai; structured outputs: ok
- [DeepSeek V4.1 Flash](https://openrouter-data-staging.dottxt.workers.dev/models/deepseek-v4.1-flash.json): `deepseek/deepseek-v4.1-flash`, by DeepSeek; structured outputs: ok
- [Hy4 preview](https://openrouter-data-staging.dottxt.workers.dev/models/hy4-preview.json): `tencent/hy4-preview`, by Tencent; no structured-output recommendation: fewer than two providers at 1,000 structured outputs a day or more
- [DeepSeek V4 Flash 0731](https://openrouter-data-staging.dottxt.workers.dev/models/deepseek-v4-flash-0731.json): `deepseek/deepseek-v4-flash-0731`, by DeepSeek; structured outputs: ok
- [GLM 5.3](https://openrouter-data-staging.dottxt.workers.dev/models/glm-5.3.json): `z-ai/glm-5.3`, by Z.ai; structured outputs: ok
- [MiMo-V2.6-Flash](https://openrouter-data-staging.dottxt.workers.dev/models/mimo-v2.6-flash.json): `xiaomi/mimo-v2.6-flash`, by Xiaomi; structured outputs: ok
- [Hy3](https://openrouter-data-staging.dottxt.workers.dev/models/hy3.json): `tencent/hy3`, by Tencent; structured outputs: ok
- [DeepSeek V4 Flash 0423](https://openrouter-data-staging.dottxt.workers.dev/models/deepseek-v4-flash.json): `deepseek/deepseek-v4-flash`, by DeepSeek; structured outputs: ok
- [MiniMax M3.1 Flash](https://openrouter-data-staging.dottxt.workers.dev/models/minimax-m3.1-flash.json): `minimax/minimax-m3.1-flash`, by MiniMax; structured outputs: ok
- [Kimi K3](https://openrouter-data-staging.dottxt.workers.dev/models/kimi-k3.json): `moonshotai/kimi-k3`, by MoonshotAI; structured outputs: ok
- [GLM 5.2](https://openrouter-data-staging.dottxt.workers.dev/models/glm-5.2.json): `z-ai/glm-5.2`, by Z.ai; structured outputs: ok
- [MiMo-V2.6-Pro](https://openrouter-data-staging.dottxt.workers.dev/models/mimo-v2.6-pro.json): `xiaomi/mimo-v2.6-pro`, by Xiaomi; structured outputs: ok
- [MiniMax M3](https://openrouter-data-staging.dottxt.workers.dev/models/minimax-m3.json): `minimax/minimax-m3`, by MiniMax; structured outputs: ok
- [Qwen3.8 27B](https://openrouter-data-staging.dottxt.workers.dev/models/qwen3.8-27b.json): `qwen/qwen3.8-27b`, by Qwen; structured outputs: ok
- [DeepSeek V4 Pro 0813](https://openrouter-data-staging.dottxt.workers.dev/models/deepseek-v4-pro-0813.json): `deepseek/deepseek-v4-pro-0813`, by DeepSeek; structured outputs: ok
- [DeepSeek V4 Pro 0423](https://openrouter-data-staging.dottxt.workers.dev/models/deepseek-v4-pro.json): `deepseek/deepseek-v4-pro`, by DeepSeek; structured outputs: ok
- [gpt-oss-120b](https://openrouter-data-staging.dottxt.workers.dev/models/gpt-oss-120b.json): `openai/gpt-oss-120b`, by OpenAI; structured outputs: ok
- [gpt-oss-20b](https://openrouter-data-staging.dottxt.workers.dev/models/gpt-oss-20b.json): `openai/gpt-oss-20b`, by OpenAI; structured outputs: ok
- [Qwen3.5-9B](https://openrouter-data-staging.dottxt.workers.dev/models/qwen3.5-9b.json): `qwen/qwen3.5-9b`, by Qwen; structured outputs: ok
- [Gemma 4 31B](https://openrouter-data-staging.dottxt.workers.dev/models/gemma-4-31b-it.json): `google/gemma-4-31b-it`, by Google; structured outputs: ok
- [Gemma 4 26B A4B](https://openrouter-data-staging.dottxt.workers.dev/models/gemma-4-26b-a4b-it.json): `google/gemma-4-26b-a4b-it`, by Google; structured outputs: ok
- [Nemotron 3.5 Super VL](https://openrouter-data-staging.dottxt.workers.dev/models/nemotron-3.5-super-vl.json): `nvidia/nemotron-3.5-super-vl`, by NVIDIA; structured outputs: ok
- [Kimi K2.6](https://openrouter-data-staging.dottxt.workers.dev/models/kimi-k2.6.json): `moonshotai/kimi-k2.6`, by MoonshotAI; structured outputs: ok
- [Qwen3.6 35B A3B](https://openrouter-data-staging.dottxt.workers.dev/models/qwen3.6-35b-a3b.json): `qwen/qwen3.6-35b-a3b`, by Qwen; structured outputs: ok
- [Beam 1.0 501B A23B](https://openrouter-data-staging.dottxt.workers.dev/models/beam-1.0-501b-a23b.json): `reflection/beam-1.0-501b-a23b`, by Reflection; structured outputs: ok
- [Qwen3 235B A22B Instruct 2507](https://openrouter-data-staging.dottxt.workers.dev/models/qwen3-235b-a22b-2507.json): `qwen/qwen3-235b-a22b-2507`, by Qwen; structured outputs: ok
- [Nemotron 3.5 Lightning](https://openrouter-data-staging.dottxt.workers.dev/models/nemotron-3.5-lightning.json): `nvidia/nemotron-3.5-lightning`, by NVIDIA; structured outputs: ok
- [DeepSeek V3.2](https://openrouter-data-staging.dottxt.workers.dev/models/deepseek-v3.2.json): `deepseek/deepseek-v3.2`, by DeepSeek; structured outputs: ok
- [Muse Spark Open](https://openrouter-data-staging.dottxt.workers.dev/models/muse-spark-open.json): `meta/muse-spark-open`, by Meta; structured outputs: ok
- [DeepSeek V4 Flash Vision Exp](https://openrouter-data-staging.dottxt.workers.dev/models/deepseek-v4-flash-vision-exp.json): `deepseek/deepseek-v4-flash-vision-exp`, by DeepSeek; no structured-output recommendation: fewer than two providers at 1,000 structured outputs a day or more
- [Llama 3.3 70B Instruct](https://openrouter-data-staging.dottxt.workers.dev/models/llama-3.3-70b-instruct.json): `meta-llama/llama-3.3-70b-instruct`, by Meta; structured outputs: ok
- [Kimi K2.7 Code](https://openrouter-data-staging.dottxt.workers.dev/models/kimi-k2.7-code.json): `moonshotai/kimi-k2.7-code`, by MoonshotAI; no structured-output recommendation: fewer than two providers at 1,000 structured outputs a day or more
- [Qwen3 30B A3B Instruct 2507](https://openrouter-data-staging.dottxt.workers.dev/models/qwen3-30b-a3b-instruct-2507.json): `qwen/qwen3-30b-a3b-instruct-2507`, by Qwen; structured outputs: ok
- [Nemotron 3 Ultra](https://openrouter-data-staging.dottxt.workers.dev/models/nemotron-3-ultra-550b-a55b.json): `nvidia/nemotron-3-ultra-550b-a55b`, by NVIDIA; no structured-output recommendation: fewer than two providers at 1,000 structured outputs a day or more
- [MiniMax M2.7](https://openrouter-data-staging.dottxt.workers.dev/models/minimax-m2.7.json): `minimax/minimax-m2.7`, by MiniMax; no structured-output recommendation: fewer than two providers at 1,000 structured outputs a day or more
- [Qwen3.5 397B A17B](https://openrouter-data-staging.dottxt.workers.dev/models/qwen3.5-397b-a17b.json): `qwen/qwen3.5-397b-a17b`, by Qwen; structured outputs: ok
- [Qwen3.8 2.4T A95B](https://openrouter-data-staging.dottxt.workers.dev/models/qwen3.8-2.4t-a95b.json): `qwen/qwen3.8-2.4t-a95b`, by Qwen; structured outputs: ok
- [Kimi K2.5](https://openrouter-data-staging.dottxt.workers.dev/models/kimi-k2.5.json): `moonshotai/kimi-k2.5`, by MoonshotAI; structured outputs: ok
- [GLM 5.1](https://openrouter-data-staging.dottxt.workers.dev/models/glm-5.1.json): `z-ai/glm-5.1`, by Z.ai; structured outputs: ok
- [MiMo-V2.5-Pro](https://openrouter-data-staging.dottxt.workers.dev/models/mimo-v2.5-pro.json): `xiaomi/mimo-v2.5-pro`, by Xiaomi; no structured-output recommendation: fewer than two providers at 1,000 structured outputs a day or more
- [GLM 5](https://openrouter-data-staging.dottxt.workers.dev/models/glm-5.json): `z-ai/glm-5`, by Z.ai; no structured-output recommendation: every provider breaks 3% of structured outputs or more
- [Llama 3.1 8B Instruct](https://openrouter-data-staging.dottxt.workers.dev/models/llama-3.1-8b-instruct.json): `meta-llama/llama-3.1-8b-instruct`, by Meta; no structured-output recommendation: fewer than two providers at 1,000 structured outputs a day or more
- [DeepSeek V3.1](https://openrouter-data-staging.dottxt.workers.dev/models/deepseek-chat-v3.1.json): `deepseek/deepseek-chat-v3.1`, by DeepSeek; structured outputs: ok
- [Granite 4.2 8B](https://openrouter-data-staging.dottxt.workers.dev/models/granite-4.2-8b.json): `ibm-granite/granite-4.2-8b`, by IBM; no structured-output recommendation: fewer than two providers at 1,000 structured outputs a day or more
- [Mistral Nemo](https://openrouter-data-staging.dottxt.workers.dev/models/mistral-nemo.json): `mistralai/mistral-nemo`, by Mistral; structured outputs: ok
- [GLM 4.7](https://openrouter-data-staging.dottxt.workers.dev/models/glm-4.7.json): `z-ai/glm-4.7`, by Z.ai; no structured-output recommendation: fewer than two providers at 1,000 structured outputs a day or more
- [Gemma 3 27B](https://openrouter-data-staging.dottxt.workers.dev/models/gemma-3-27b-it.json): `google/gemma-3-27b-it`, by Google; no recommendation: fewer than two providers at 1,000 tool calls a day or more; structured outputs: ok

## Providers

Each provider's page: its tool-call error rate on every model it serves, best first, and its rank among each model's providers. The same numbers are in the models' files, under the provider's `id`.

- [AkashML](https://openrouter-data-staging.dottxt.workers.dev/p/akashml/): 0.52% to 32.75% of tool calls broken across 8 models
- [Alibaba Cloud Int.](https://openrouter-data-staging.dottxt.workers.dev/p/alibaba/): 0.05% to 15.01% of tool calls broken across 14 models
- [Amazon Bedrock](https://openrouter-data-staging.dottxt.workers.dev/p/amazon-bedrock/): 0.05% to 6.65% of tool calls broken across 5 models
- [AtlasCloud](https://openrouter-data-staging.dottxt.workers.dev/p/atlas-cloud/): 0.30% to 36.08% of tool calls broken across 16 models
- [Azure](https://openrouter-data-staging.dottxt.workers.dev/p/azure/): 4.07% to 5.81% of tool calls broken across 2 models
- [Baidu Qianfan](https://openrouter-data-staging.dottxt.workers.dev/p/baidu/): 0.50% to 6.20% of tool calls broken across 10 models
- [Baseten](https://openrouter-data-staging.dottxt.workers.dev/p/baseten/): 0.06% to 9.87% of tool calls broken across 11 models
- [Cerebras](https://openrouter-data-staging.dottxt.workers.dev/p/cerebras/): 13.98% to 36.42% of tool calls broken across 2 models
- [Chutes](https://openrouter-data-staging.dottxt.workers.dev/p/chutes/): 0.87% to 2.07% of tool calls broken across 4 models
- [Cloudflare](https://openrouter-data-staging.dottxt.workers.dev/p/cloudflare/): 0.40% to 7.93% of tool calls broken across 10 models
- [Cohere](https://openrouter-data-staging.dottxt.workers.dev/p/cohere/): 0.17% to 2.70% of tool calls broken across 6 models
- [CoreWeave](https://openrouter-data-staging.dottxt.workers.dev/p/coreweave/): 0.29% to 67.14% of tool calls broken across 20 models
- [Crusoe](https://openrouter-data-staging.dottxt.workers.dev/p/crusoe/): 0.46% to 7.33% of tool calls broken across 10 models
- [Darkbloom](https://openrouter-data-staging.dottxt.workers.dev/p/darkbloom/): 0.00% to 0.13% of tool calls broken across 5 models
- [Decart](https://openrouter-data-staging.dottxt.workers.dev/p/decart/): 0.34% to 1.47% of tool calls broken across 5 models
- [DeepInfra](https://openrouter-data-staging.dottxt.workers.dev/p/deepinfra/): 0.01% to 9.96% of tool calls broken across 33 models
- [DeepSeek](https://openrouter-data-staging.dottxt.workers.dev/p/deepseek/): 1.05% to 2.01% of tool calls broken across 2 models
- [DekaLLM](https://openrouter-data-staging.dottxt.workers.dev/p/dekallm/): 0.40% to 3.94% of tool calls broken across 6 models
- [DigitalOcean](https://openrouter-data-staging.dottxt.workers.dev/p/digitalocean/): 0.20% to 3.81% of tool calls broken across 13 models
- [Fireworks](https://openrouter-data-staging.dottxt.workers.dev/p/fireworks/): 0.01% to 11.02% of tool calls broken across 8 models
- [Friendli](https://openrouter-data-staging.dottxt.workers.dev/p/friendli/): 0.00% to 0.80% of tool calls broken across 5 models
- [GMICloud](https://openrouter-data-staging.dottxt.workers.dev/p/gmicloud/): 0.58% to 23.02% of tool calls broken across 17 models
- [Google Vertex](https://openrouter-data-staging.dottxt.workers.dev/p/google-vertex/): 0.50% to 5.30% of tool calls broken across 2 models
- [Groq](https://openrouter-data-staging.dottxt.workers.dev/p/groq/): 0.04% to 2.21% of tool calls broken across 5 models
- [Inceptron](https://openrouter-data-staging.dottxt.workers.dev/p/inceptron/): 0.22% to 0.56% of tool calls broken across 6 models
- [inference.net](https://openrouter-data-staging.dottxt.workers.dev/p/inference-net/): 0.06% to 1.74% of tool calls broken across 6 models
- [io.net](https://openrouter-data-staging.dottxt.workers.dev/p/io-net/): 0.86% to 4.73% of tool calls broken across 6 models
- [Ionstream](https://openrouter-data-staging.dottxt.workers.dev/p/ionstream/): 0.04% to 1.49% of tool calls broken across 3 models
- [Makora](https://openrouter-data-staging.dottxt.workers.dev/p/makora/): 0.32% to 3.92% of tool calls broken across 8 models
- [MiniMax](https://openrouter-data-staging.dottxt.workers.dev/p/minimax/): 1.76% to 2.98% of tool calls broken across 3 models
- [Mistral](https://openrouter-data-staging.dottxt.workers.dev/p/mistral/): 0.62% to 5.14% of tool calls broken across 4 models
- [Modal](https://openrouter-data-staging.dottxt.workers.dev/p/modal/): 0.20% to 3.26% of tool calls broken across 7 models
- [ModelRun [by Modular]](https://openrouter-data-staging.dottxt.workers.dev/p/modelrun/): 0.09% to 1.42% of tool calls broken across 5 models
- [Moonshot AI](https://openrouter-data-staging.dottxt.workers.dev/p/moonshotai/): 0.22% to 2.92% of tool calls broken across 3 models
- [Morph](https://openrouter-data-staging.dottxt.workers.dev/p/morph/): 0.07% to 0.44% of tool calls broken across 6 models
- [NEAR AI](https://openrouter-data-staging.dottxt.workers.dev/p/near-ai/): 0.94% to 10.37% of tool calls broken across 3 models
- [Nebius Token Factory](https://openrouter-data-staging.dottxt.workers.dev/p/nebius/): 0.07% to 80.12% of tool calls broken across 12 models
- [NextBit](https://openrouter-data-staging.dottxt.workers.dev/p/nextbit/): 0.00% to 0.00% of tool calls broken across 2 models
- [NovitaAI](https://openrouter-data-staging.dottxt.workers.dev/p/novita/): 0.03% to 29.28% of tool calls broken across 26 models
- [Open Inference](https://openrouter-data-staging.dottxt.workers.dev/p/open-inference/): 0.32% to 1.46% of tool calls broken across 3 models
- [Parasail](https://openrouter-data-staging.dottxt.workers.dev/p/parasail/): 0.21% to 22.66% of tool calls broken across 19 models
- [Particle AI](https://openrouter-data-staging.dottxt.workers.dev/p/particle-ai/): 0.23% to 0.34% of tool calls broken across 2 models
- [Phala](https://openrouter-data-staging.dottxt.workers.dev/p/phala/): 0.17% to 11.30% of tool calls broken across 12 models
- [Prime Intellect](https://openrouter-data-staging.dottxt.workers.dev/p/primeintellect/): 0.02% to 0.58% of tool calls broken across 6 models
- [Reka AI](https://openrouter-data-staging.dottxt.workers.dev/p/reka/): 0.18% to 10.07% of tool calls broken across 6 models
- [Relace](https://openrouter-data-staging.dottxt.workers.dev/p/relace/): 0.05% to 4.17% of tool calls broken across 11 models
- [Sail Research](https://openrouter-data-staging.dottxt.workers.dev/p/sail-research/): 0.03% to 1.06% of tool calls broken across 7 models
- [SambaNova](https://openrouter-data-staging.dottxt.workers.dev/p/sambanova/): 2.53% to 5.05% of tool calls broken across 3 models
- [SiliconFlow](https://openrouter-data-staging.dottxt.workers.dev/p/siliconflow/): 0.06% to 89.86% of tool calls broken across 22 models
- [StreamLake](https://openrouter-data-staging.dottxt.workers.dev/p/streamlake/): 0.09% to 6.96% of tool calls broken across 14 models
- [Tencent Cloud](https://openrouter-data-staging.dottxt.workers.dev/p/tencent/): 0.16% to 1.46% of tool calls broken across 2 models
- [Tenstorrent](https://openrouter-data-staging.dottxt.workers.dev/p/tenstorrent/): 1.19% to 2.77% of tool calls broken across 2 models
- [Together](https://openrouter-data-staging.dottxt.workers.dev/p/together/): 0.25% to 4.19% of tool calls broken across 15 models
- [Venice](https://openrouter-data-staging.dottxt.workers.dev/p/venice/): 0.19% to 17.23% of tool calls broken across 19 models
- [Wafer](https://openrouter-data-staging.dottxt.workers.dev/p/wafer/): 0.12% to 1.06% of tool calls broken across 10 models
- [Xiaomi](https://openrouter-data-staging.dottxt.workers.dev/p/xiaomi/): 1.71% to 4.59% of tool calls broken across 2 models
- [Z.ai](https://openrouter-data-staging.dottxt.workers.dev/p/z-ai/): 0.65% to 2.22% of tool calls broken across 6 models

## Agent resources

- [Routing guide](https://openrouter-data-staging.dottxt.workers.dev/routing-guide.md): How to check the evidence and present routing choices to the user.
- [Data schema](https://openrouter-data-staging.dottxt.workers.dev/schema.json): Machine-readable definition of every file.
- [Human-readable index](https://openrouter-data-staging.dottxt.workers.dev/index.md): Overview, findings, and how the headline numbers are computed.

## Optional

- [Every model](https://openrouter-data-staging.dottxt.workers.dev/data.json): every model's file in one document; large.
- [Snapshots](https://openrouter-data-staging.dottxt.workers.dev/snapshots/index.json): each day's data, `snapshots/{date}.json`.