{
  "listings": [
    {
      "slug": "cloudflare-clef",
      "name": "Clef",
      "company": "Cloudflare, Inc.",
      "grade": "B",
      "fit": "yes",
      "why": "Images on Workers AI (up to 4 per request, 16 megapixels each) and video through the weights or Clef-omni; Clef-flash median 38.8 ms, Clef 209 ms. No option-order control.",
      "source": "https://developers.cloudflare.com/workers-ai/models/clef/",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/cloudflare-clef"
    },
    {
      "slug": "decision-3",
      "name": "Decision 3.0",
      "company": "vLLM Semantic Router project",
      "grade": "B",
      "fit": "yes",
      "why": "Images (several, up to 1.6 megapixels) and video at 2 frames a second; d3-mini answers an image request in 112 ms median, d3 in 290 ms; permutation_average counters first-option bias.",
      "source": "https://huggingface.co/vllm-sr/d3-mini",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/decision-3"
    },
    {
      "slug": "liquid-d1",
      "name": "Liquid d1",
      "company": "Liquid AI, Inc.",
      "grade": "C",
      "fit": "yes",
      "why": "Up to 8 base64 images a request on the hosted API at 0.20 s p50 (Arize); d1-2B adds video and runs on MLX and llama.cpp locally.",
      "source": "https://docs.liquid.ai/lfm/models/d1",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/liquid-d1"
    },
    {
      "slug": "openai-decisions-api",
      "name": "OpenAI Decisions API",
      "company": "OpenAI OpCo, LLC",
      "grade": "A",
      "fit": "yes",
      "why": "Images as base64 or public URLs, conversation and tool outputs in input, 0.16 s p50 at Arize, and a voice guide for GPT-Live action selection. No option-order control.",
      "source": "https://developers.openai.com/api/docs/guides/decisions",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/openai-decisions-api"
    },
    {
      "slug": "pplx-decider",
      "name": "pplx-decider",
      "company": "Perplexity AI, Inc.",
      "grade": "BB",
      "fit": "yes",
      "why": "Up to 8 base64 images a question at 2,048 tiles each, a browser-agent cookbook asking six questions a screenshot; images over the tile cap time out at 504 rather than failing fast.",
      "source": "https://docs.perplexity.ai/docs/decisions/quickstart",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/pplx-decider"
    },
    {
      "slug": "pplx-decider-flash",
      "name": "pplx-decider-flash",
      "company": "Perplexity AI, Inc.",
      "grade": "B",
      "fit": "yes",
      "why": "Images up to 8 a question on the hosted API and the Mac app; the 4B is built for latency, and the Mac app's examples label screenshots locally.",
      "source": "https://docs.perplexity.ai/docs/decisions/quickstart",
      "confidence": "medium",
      "url": "https://www.anchorterminal.com/tools/pplx-decider-flash"
    },
    {
      "slug": "celeris-1-decision",
      "name": "Celeris-1 Decision",
      "company": "Marqo Inc",
      "grade": "D",
      "fit": "caveat",
      "why": "Up to 8 images a request at 25 megapixels, read in order as image_1, image_2. No published latency for the decision model and no control for option order.",
      "source": "https://docs.celeris.ai/decisions.md",
      "confidence": "medium",
      "url": "https://www.anchorterminal.com/tools/celeris-1-decision"
    },
    {
      "slug": "convai-laya",
      "name": "Laya",
      "company": "Convai Innovations Pvt. Ltd.",
      "grade": "BB",
      "fit": "caveat",
      "why": "Fast enough at 33 ms on a GPU, but the core checkpoints read text only (images route to the optional vega engine) and issue 131 shows a position bias on score options.",
      "source": "https://github.com/NandhaKishorM/laya",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/convai-laya"
    },
    {
      "slug": "decider",
      "name": "Decider",
      "company": "Mark Marosi (Mapika)",
      "grade": "BB",
      "fit": "caveat",
      "why": "Images in /v1/systemone since 1.10.0 on the CUDA Python path (GGUF is text only); the README's browser results are 22 click-only MiniWoB++ tasks and issue 9 shows a destructive click. 133 ms on MPS.",
      "source": "https://github.com/Mapika/decider#readme",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/decider"
    },
    {
      "slug": "drex-max-v1",
      "name": "Drex Max v1",
      "company": "Nace AI, Inc.",
      "grade": "C",
      "fit": "caveat",
      "why": "Images need the Python loader on an NVIDIA GPU (up to 4,194,304 pixels each); the GGUF, Q4, Kapitsa and MLX paths read text only. Public vision index 73.92 is the vendor's run. No latency figure for image requests.",
      "source": "https://huggingface.co/nace-ai/drex-max-v1",
      "confidence": "medium",
      "url": "https://www.anchorterminal.com/tools/drex-max-v1"
    },
    {
      "slug": "elephant-1",
      "name": "elephant-1",
      "company": "Baseten Labs, Inc.",
      "grade": "D",
      "fit": "caveat",
      "why": "Images are accepted, but the report says the vision encoder is frozen and training ran on text alone, so image decisions are untested by the authors. No per-decision latency on the card.",
      "source": "https://huggingface.co/baselabs/elephant-1",
      "confidence": "medium",
      "url": "https://www.anchorterminal.com/tools/elephant-1"
    },
    {
      "slug": "jaredpalmer-kev",
      "name": "Kev",
      "company": "Jared Palmer",
      "grade": "BB",
      "fit": "caveat",
      "why": "Text only, no image input; a chess demo and the permute endpoint cover order bias, and Kev-4B answers in 18 ms on an H100. Screenshot environments need a text description.",
      "source": "https://github.com/jaredpalmer/kev#readme",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/jaredpalmer-kev"
    },
    {
      "slug": "jev-27b-vl",
      "name": "JEV-27B-VL",
      "company": "AutoTrust AI Lab",
      "grade": "C",
      "fit": "caveat",
      "why": "Reads screenshots and camera images: about 150 to 300 ms a decision in the Mario, Tetris and robot-arm demos and 239 ms a robot-arm step, on an 80 GB GPU. Over the 200 ms local target and no order control.",
      "source": "https://huggingface.co/autotrust/JEV-27B-VL",
      "confidence": "medium",
      "url": "https://www.anchorterminal.com/tools/jev-27b-vl"
    },
    {
      "slug": "joey-1",
      "name": "joey-1",
      "company": "Baseten Labs, Inc.",
      "grade": "C",
      "fit": "caveat",
      "why": "Images are accepted, but the report says the vision encoder is frozen and training ran on text alone, so image decisions are untested by the authors. No per-decision latency on the card.",
      "source": "https://huggingface.co/baselabs/joey-1",
      "confidence": "medium",
      "url": "https://www.anchorterminal.com/tools/joey-1"
    },
    {
      "slug": "mammoth-1",
      "name": "mammoth-1",
      "company": "Baseten",
      "grade": "D",
      "fit": "caveat",
      "why": "Images are accepted, but the report says the vision encoder is frozen and training ran on text alone, so image decisions are untested by the authors. No per-decision latency on the card.",
      "source": "https://huggingface.co/baselabs/mammoth-1",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/mammoth-1"
    },
    {
      "slug": "nace-drex",
      "name": "Drex 1.5",
      "company": "Nace AI, Inc.",
      "grade": "BB",
      "fit": "caveat",
      "why": "Images (up to 8 PNG or JPEG) are accepted on the Drex Max models only, as an images array or the TypeSafe multipart form. No latency figure for image requests.",
      "source": "https://console.nace.ai/openapi.json",
      "confidence": "medium",
      "url": "https://www.anchorterminal.com/tools/nace-drex"
    },
    {
      "slug": "rightnow-system-zero",
      "name": "RightNow System Zero",
      "company": "RightNow AI, Inc.",
      "grade": "B",
      "fit": "caveat",
      "why": "Images, 16 kHz WAV audio, PDFs and preview video as inline input, each with its own file fee (USD 0.011 an image); a measured latency table exists but no guarantee. Uncalibrated probabilities.",
      "source": "https://www.rightnowai.co/llms-full.txt",
      "confidence": "medium",
      "url": "https://www.anchorterminal.com/tools/rightnow-system-zero"
    },
    {
      "slug": "sprag-spev",
      "name": "SPEV",
      "company": "Sprag AI, Inc.",
      "grade": "D",
      "fit": "caveat",
      "why": "One inline image per request plus audio and video; no latency figure, one image means no before-and-after comparison.",
      "source": "https://sprag.ai/docs/decisions/api",
      "confidence": "medium",
      "url": "https://www.anchorterminal.com/tools/sprag-spev"
    },
    {
      "slug": "sqwish-kettei",
      "name": "kettei",
      "company": "Sqwish Ltd, trading as Sqwish Labs",
      "grade": "C",
      "fit": "caveat",
      "why": "kettei-chub and kettei-fat read up to 8 pictures with the context; the default kettei-core is text only and no latency is published.",
      "source": "https://docs.sqwish.ai/concepts",
      "confidence": "medium",
      "url": "https://www.anchorterminal.com/tools/sqwish-kettei"
    },
    {
      "slug": "strands-decider",
      "name": "Strands Decider 2B",
      "company": "Amazon Web Services, Inc.",
      "grade": "B",
      "fit": "caveat",
      "why": "Text first; the Qwen3.5 v1 model answers over images zero-shot with the vision extra (Gemma 4 models are text only), 115 ms on a GPU, and the 55 percent first-answer preference measured by AWS Builder.",
      "source": "https://github.com/strands-labs/strands-decider#readme",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/strands-decider"
    },
    {
      "slug": "typesafe-jev",
      "name": "Jev",
      "company": "TypeSafe AI, Inc.",
      "grade": "B",
      "fit": "caveat",
      "why": "Images via the TypeSafe-Danger-Zone-Images header and replies under 400 ms through OpenRouter, but the jaggedness page records a lean to the first option and users saw Jev keep picking the first move in a six-room map and fail basic mazes.",
      "source": "https://docs.typesafe.ai/model-jaggedness/jev-1.13",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/typesafe-jev"
    },
    {
      "slug": "gliclass",
      "name": "GLiClass",
      "company": "Knowledgator Engineering Ltd.",
      "grade": "C",
      "fit": "no",
      "why": "Text only; no image input in the pipeline or server.",
      "source": "https://docs.knowledgator.com/docs/frameworks/gliclass/usage/",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/gliclass"
    },
    {
      "slug": "granite-switch",
      "name": "Granite Switch 4.1",
      "company": "International Business Machines Corporation",
      "grade": "B",
      "fit": "no",
      "why": "Text and audio only; the model pages document no image input.",
      "source": "https://huggingface.co/ibm-granite/granite-switch-4.1-8b",
      "confidence": "medium",
      "url": "https://www.anchorterminal.com/tools/granite-switch"
    },
    {
      "slug": "microsoft-decision-1",
      "name": "Microsoft-Decision-1",
      "company": "Microsoft",
      "grade": "E",
      "fit": "no",
      "why": "Text and JSON input only; OpenRouter lists text as the sole input modality.",
      "source": "https://openrouter.ai/microsoft/microsoft-decision-1",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/microsoft-decision-1"
    },
    {
      "slug": "scaledown",
      "name": "ScaleDown",
      "company": "Carbon ScaleDown, Inc.",
      "grade": "E",
      "fit": "no",
      "why": "Images and PDFs are OCR-extracted to text before classification, so the model never reads pixels; no screenshot decisions.",
      "source": "https://docs.scaledown.ai/api-reference/classify",
      "confidence": "medium",
      "url": "https://www.anchorterminal.com/tools/scaledown"
    },
    {
      "slug": "vela",
      "name": "Vela 2.0",
      "company": "vLLM Semantic Router project (with KR Labs)",
      "grade": "BB",
      "fit": "no",
      "why": "Text only; the card documents no image input.",
      "source": "https://huggingface.co/vllm-sr/Vela-2.0-4B",
      "confidence": "high",
      "url": "https://www.anchorterminal.com/tools/vela"
    }
  ],
  "marks": "yes: the docs show it does the job as the success test says. caveat: it does the job with a limit an agent should know. no: it doesn't do this job. unchecked: we couldn't read it.",
  "meta": {
    "attribution": "Anchor Terminal (https://www.anchorterminal.com)",
    "attributionRequired": true,
    "credit": "CC BY 4.0, credit Anchor Terminal and link the page or API address you took it from (https://www.anchorterminal.com/terms/#data)",
    "docs": "https://www.anchorterminal.com/docs/",
    "generatedAt": "2026-11-16",
    "license": "CC-BY-4.0",
    "method": "https://www.anchorterminal.com/benchmark/",
    "methodology": "0.8",
    "openapi": "https://www.anchorterminal.com/openapi.json",
    "preview": false,
    "run": "2026-10-01",
    "runLabel": "October 2026 research run",
    "terms": "https://www.anchorterminal.com/terms/#data"
  },
  "useCase": {
    "accounts": 6,
    "after": [
      "browser-automation.click",
      "computer-use.act"
    ],
    "area": "models",
    "asks": [
      "which direction should the agent move",
      "click, wait, retry or stop",
      "which link on this page leads to the target",
      "is the game state winning"
    ],
    "before": [
      "browser-automation.screenshot",
      "computer-use.read-screen"
    ],
    "blockers": [
      "first-option bias stalls exploration",
      "models fail multi-step reasoning such as mazes",
      "images need a CUDA loader on several open models"
    ],
    "capabilities": [
      "inference.decision"
    ],
    "category": "decision-models",
    "categoryName": "Decision models",
    "checked": "2026-11-12",
    "companies": {
      "fit": 5,
      "fitWithLimit": 13,
      "tagged": 30
    },
    "competition": "crowded",
    "fits": [
      "cloudflare-clef",
      "decision-3",
      "liquid-d1",
      "openai-decisions-api",
      "pplx-decider",
      "pplx-decider-flash"
    ],
    "fitsWithLimit": [
      "celeris-1-decision",
      "convai-laya",
      "decider",
      "drex-max-v1",
      "elephant-1",
      "jaredpalmer-kev",
      "jev-27b-vl",
      "joey-1",
      "mammoth-1",
      "nace-drex",
      "rightnow-system-zero",
      "sprag-spev",
      "sqwish-kettei",
      "strands-decider",
      "typesafe-jev"
    ],
    "id": "decision-models.agent-step-choice",
    "listings": {
      "caveat": 15,
      "no": 5,
      "notRead": 1,
      "tagged": 33,
      "unchecked": 0,
      "yes": 6
    },
    "name": "Choose the next step for an agent acting in a UI, game or environment",
    "status": "draft",
    "success": "A move per call, chosen from the offered options. Measured: each move in under 400 ms hosted or 200 ms local, the episode completed at least as often as a rule baseline, without always picking the first option.",
    "url": "https://www.anchorterminal.com/api/v1/use-cases/decision-models.agent-step-choice.json",
    "weight": 7,
    "weightBasis": "6 of 41 accounts, all small experiments (mazes, rooms, games, form filling); weighted below the business uses because none was in production"
  }
}
