{
  "version": 2,
  "updated": "2026-08-06",
  "note": "RapidCoder curated model catalog. Host at https://rapiddashboard.ai/rapid.coder/model.json with an 'Access-Control-Allow-Origin: *' header or the browser cross-origin fetch is blocked. Model ids are OpenRouter slugs and must match a live model on openrouter.ai/models or they are skipped. cost is USD per 1M tokens (input/output) pulled from OpenRouter on the 'updated' date. The 'human_eval' and 'value' blocks are additive as of 2026-08-06 and safe to ignore. Arena leaderboard suffixes like -max, -high and -xhigh are reasoning-effort settings, not separate models, so they collapse onto the base OpenRouter slug here.",
  "default": {
    "provider": "openrouter",
    "model": "z-ai/glm-5.2"
  },
  "curatedOnly": true,
  "inclusionPolicy": {
    "summary": "Each model carries an 'included' flag. true means inference for that model is covered by the RapidCoder subscription and billed to us; false means the user supplies their own OpenRouter key and pays the provider directly.",
    "advisory": "This flag is presentation metadata only. model.json is a public static file served with 'Access-Control-Allow-Origin: *', so a client can read it, ignore it, or fabricate its own copy. It exists so RapidCoder can label and default correctly in the UI — it is NOT an entitlement check. The authoritative decision about whether a request is billed to our key must be made server-side, at the gateway that holds the key, against the account's plan. Never let a client-supplied model id, flow config, or 'included' claim decide who pays.",
    "flags": {
      "included": "Covered by the subscription. Keep this set to low-cost models only; every model marked included is a direct cost exposure multiplied by every user on the plan.",
      "byok": "Not covered. RapidCoder must require a user-supplied key before allowing selection."
    }
  },
  "scoreModel": {
    "summary": "Every model carries two independent scores, because building an application and designing one are different jobs and the rankings genuinely diverge.",
    "code": {
      "label": "Code",
      "measures": "Functional correctness — does the application work. Code generation, debugging, refactoring, completion accuracy.",
      "source": "Artificial Analysis coding index, 0-100",
      "method": "automated test suites",
      "why": "Whether code runs is objectively checkable, so an automated benchmark is the right instrument."
    },
    "design": {
      "label": "Design",
      "measures": "Visual and UX quality — does the result look like something you would put in front of a client.",
      "source": "Design Arena mean win rate across the website, uicomponent and dataviz categories, 0-100",
      "method": "blind human pairwise voting",
      "why": "Whether a screen looks good is not testable, so only human preference measures it. The score is the percentage of blind head-to-head matchups the model won.",
      "excluded_categories": "3d, svg, asciiart and gamedev are rated by Design Arena but excluded here as irrelevant to dashboard work."
    },
    "guidance": "For most business use the Code score is the one to weight: dashboards and reports must be correct before they are attractive. Reach for the Design score when the deliverable is client-facing or presentation-grade."
  },
  "selectionMethod": "Human preference first. A model qualifies for the catalog on its standing in blind human-vote arenas, and only then do we rank the qualifiers by cost. We do not promote a model on automated benchmark scores alone: benchmarks measure whether code passes tests, arenas measure whether a person preferred the result, and for dashboard work the second question is the one that matters. Artificial Analysis indices are retained as a cost lens and a tiebreaker, not as the qualifier.",
  "costBasis": "Cost of intelligence = blended cost divided by coding_index, where blended cost = input + output USD per 1M tokens. Lower is better: it is the dollars you pay per point of coding capability. Treat it as a cost lens rather than a ranking, because its denominator is an automated benchmark. Note that dividing cost by an arena Elo instead would be meaningless: Elo has an arbitrary zero and the whole catalog sits inside a 130-point band, so the ratio would collapse into 'cheapest wins'. Use arena standing to decide which models are good enough, then use this number to choose among them.",
  "benchmarkSources": {
    "human_eval": [
      {
        "name": "WebDev Arena",
        "url": "https://arena.ai/leaderboard/code/webdev",
        "method": "Blind pairwise human voting on general web development tasks.",
        "role": "Primary qualifier for the catalog."
      },
      {
        "name": "Design Arena",
        "url": "https://designarena.ai",
        "method": "Blind pairwise human voting, Bradley-Terry ratings, scored per design category (website, uicomponent, dataviz).",
        "role": "Secondary qualifier. Narrower and more visual than WebDev Arena, and the closest available proxy for dashboard work. Consulted via OpenRouter's design_arena benchmark data."
      }
    ],
    "automated": [
      {
        "name": "Artificial Analysis",
        "url": "https://artificialanalysis.ai",
        "method": "Automated test suites producing intelligence, coding and agentic indices, 0-100.",
        "role": "Cost lens only. Accessed via OpenRouter."
      }
    ],
    "pricing": "OpenRouter live model pricing",
    "caveat": "The two human arenas do not always agree. Where they diverge, the design-specific one carries more weight for this product, because RapidDashboard output is judged by a person looking at a screen."
  },
  "groups": [
    {
      "label": "Free Models",
      "models": [
        {
          "provider": "openrouter",
          "model": "nvidia/nemotron-3-ultra-550b-a55b:free",
          "name": "Nemotron 3 Ultra (Free)",
          "tagline": "Free • 1M context",
          "tier": "free",
          "included": true,
          "cost": {
            "input": 0,
            "output": 0,
            "unit": "usd_per_million_tokens",
            "context": 1000000
          },
          "human_eval": {
            "webdev_arena": null,
            "design_arena": {
              "website": 83,
              "uicomponent": 58,
              "dataviz": 73
            }
          },
          "scores": {
            "code": {
              "value": 49.3,
              "rank": 6,
              "of": 6,
              "source": "Artificial Analysis coding index",
              "method": "automated"
            },
            "design": {
              "value": 36.6,
              "rank": 5,
              "of": 5,
              "source": "Design Arena",
              "method": "human votes",
              "category_win_rates": {
                "website": 33.0,
                "uicomponent": 39.5,
                "dataviz": 37.4
              }
            }
          },
          "benchmarks": {
            "source": "Artificial Analysis (via OpenRouter)",
            "intelligence_index": 37.8,
            "coding_index": 49.3,
            "agentic_index": 27.4
          },
          "value": {
            "blended_cost": 0,
            "usd_per_coding_point": 0,
            "rank": 0,
            "note": "Free, so cost of intelligence is undefined rather than best-in-class. Human voters rank it well down the field. Prototype here, ship elsewhere."
          },
          "description": "NVIDIA's free flagship. No cost, large context — good for everyday tasks."
        }
      ]
    },
    {
      "label": "Preferred Models",
      "models": [
        {
          "provider": "openrouter",
          "model": "deepseek/deepseek-v4-flash-0731",
          "name": "DeepSeek V4 Flash",
          "tagline": "Cheapest capable option",
          "tier": "budget",
          "included": true,
          "cost": {
            "input": 0.09,
            "output": 0.18,
            "unit": "usd_per_million_tokens",
            "context": 1048576
          },
          "human_eval": {
            "webdev_arena": {
              "rank": 8,
              "score": 1577,
              "votes": 1319,
              "preliminary": true
            },
            "design_arena": {
              "website": 29,
              "uicomponent": 37,
              "dataviz": 41
            }
          },
          "scores": {
            "code": {
              "value": 69.1,
              "rank": 4,
              "of": 6,
              "source": "Artificial Analysis coding index",
              "method": "automated"
            },
            "design": {
              "value": 45.1,
              "rank": 4,
              "of": 5,
              "source": "Design Arena",
              "method": "human votes",
              "category_win_rates": {
                "website": 48.1,
                "uicomponent": 42.6,
                "dataviz": 44.7
              }
            }
          },
          "benchmarks": {
            "source": "Artificial Analysis (via OpenRouter)",
            "intelligence_index": 49.9,
            "coding_index": 69.1,
            "agentic_index": 45.7
          },
          "value": {
            "blended_cost": 0.27,
            "usd_per_coding_point": 0.0039,
            "rank": 1,
            "note": "Roughly 12x cheaper than the default. The two human arenas disagree about it: near-tied with the default on WebDev Arena (#8 vs #7, and still preliminary on 1,319 votes), but 20-plus ranks behind on every Design Arena visual category. Bulk and repetitive work, not the dashboard someone presents."
          },
          "description": "Frontier-adjacent benchmark scores at throwaway prices. The right pick for high-volume, repetitive jobs."
        },
        {
          "provider": "openrouter",
          "model": "z-ai/glm-5.2",
          "name": "GLM 5.2",
          "tagline": "Powerful, low cost",
          "tier": "balanced",
          "included": false,
          "recommended": true,
          "cost": {
            "input": 0.76,
            "output": 2.42,
            "unit": "usd_per_million_tokens",
            "context": 1048576
          },
          "human_eval": {
            "webdev_arena": {
              "rank": 7,
              "score": 1586,
              "votes": 6361,
              "preliminary": false
            },
            "design_arena": {
              "website": 3,
              "uicomponent": 6,
              "dataviz": 6
            }
          },
          "scores": {
            "code": {
              "value": 68.8,
              "rank": 5,
              "of": 6,
              "source": "Artificial Analysis coding index",
              "method": "automated"
            },
            "design": {
              "value": 58.4,
              "rank": 3,
              "of": 5,
              "source": "Design Arena",
              "method": "human votes",
              "category_win_rates": {
                "website": 60.2,
                "uicomponent": 58.0,
                "dataviz": 57.0
              }
            }
          },
          "benchmarks": {
            "source": "Artificial Analysis (via OpenRouter)",
            "intelligence_index": 51.1,
            "coding_index": 68.8,
            "agentic_index": 43.1
          },
          "value": {
            "blended_cost": 3.18,
            "usd_per_coding_point": 0.0462,
            "rank": 2,
            "note": "The default because it is the cheapest model that human voters still rank near the top for visual work: #3 website and #6 dataviz on Design Arena. Cheaper models match it on automated coding scores but not on what people actually pick."
          },
          "description": "Strong coding at a fraction of frontier cost, and the cheapest model human voters still rate highly for web UI. The value default."
        },
        {
          "provider": "openrouter",
          "model": "qwen/qwen3.8-max",
          "name": "Qwen3.8 Max",
          "tagline": "Best agentic score per dollar",
          "tier": "balanced",
          "included": false,
          "recommended": true,
          "cost": {
            "input": 2,
            "output": 6,
            "unit": "usd_per_million_tokens",
            "context": 1000000
          },
          "human_eval": {
            "webdev_arena": {
              "rank": 4,
              "score": 1668,
              "votes": 1563,
              "preliminary": true
            },
            "design_arena": null
          },
          "scores": {
            "code": {
              "value": 71.8,
              "rank": 3,
              "of": 6,
              "source": "Artificial Analysis coding index",
              "method": "automated"
            },
            "design": {
              "value": null,
              "rank": null,
              "source": "Design Arena",
              "method": "human votes",
              "note": "Not yet rated by Design Arena — too new to have accumulated votes."
            }
          },
          "benchmarks": {
            "source": "Artificial Analysis (via OpenRouter)",
            "intelligence_index": 56.2,
            "coding_index": 71.8,
            "agentic_index": 55.4
          },
          "value": {
            "blended_cost": 8,
            "usd_per_coding_point": 0.1114,
            "rank": 3,
            "note": "#4 on WebDev Arena, though still preliminary on 1,563 votes and not yet rated by Design Arena. Highest agentic index in the catalog — above Claude Opus 5 — at roughly a quarter of the blended cost."
          },
          "description": "The long-run workhorse. Best choice for overnight app builds and multi-step integrations."
        },
        {
          "provider": "openrouter",
          "model": "moonshotai/kimi-k3",
          "name": "Kimi K3",
          "tagline": "Frontier coding, mid-tier price",
          "tier": "premium",
          "included": false,
          "recommended": true,
          "cost": {
            "input": 3,
            "output": 15,
            "unit": "usd_per_million_tokens",
            "context": 1048576
          },
          "human_eval": {
            "webdev_arena": {
              "rank": 2,
              "score": 1676,
              "votes": 4366,
              "preliminary": false
            },
            "design_arena": {
              "website": 1,
              "uicomponent": 1,
              "dataviz": 2
            }
          },
          "scores": {
            "code": {
              "value": 76.2,
              "rank": 2,
              "of": 6,
              "source": "Artificial Analysis coding index",
              "method": "automated"
            },
            "design": {
              "value": 64.6,
              "rank": 1,
              "of": 5,
              "source": "Design Arena",
              "method": "human votes",
              "category_win_rates": {
                "website": 63.5,
                "uicomponent": 64.8,
                "dataviz": 65.4
              }
            }
          },
          "benchmarks": {
            "source": "Artificial Analysis (via OpenRouter)",
            "intelligence_index": 57.1,
            "coding_index": 76.2,
            "agentic_index": 50.1
          },
          "value": {
            "blended_cost": 18,
            "usd_per_coding_point": 0.2362,
            "rank": 4,
            "note": "The rare model both arenas agree on: #2 on WebDev Arena and #1 for website and UI components on Design Arena, at 60% of frontier price. Best human-rated web quality per dollar."
          },
          "description": "Near-frontier coding at 60% of frontier price, and the top human-rated model for building web interfaces."
        },
        {
          "provider": "openrouter",
          "model": "anthropic/claude-opus-5",
          "name": "Claude Opus 5",
          "tagline": "Frontier — top quality",
          "tier": "frontier",
          "included": false,
          "cost": {
            "input": 5,
            "output": 25,
            "unit": "usd_per_million_tokens",
            "context": 1000000
          },
          "human_eval": {
            "webdev_arena": {
              "rank": 1,
              "score": 1705,
              "votes": 2192,
              "preliminary": false,
              "note": "#1 at max reasoning effort, #3 at high effort — the same model at two settings."
            },
            "design_arena": {
              "website": 2,
              "uicomponent": 2,
              "dataviz": 1
            }
          },
          "scores": {
            "code": {
              "value": 78.0,
              "rank": 1,
              "of": 6,
              "source": "Artificial Analysis coding index",
              "method": "automated"
            },
            "design": {
              "value": 62.0,
              "rank": 2,
              "of": 5,
              "source": "Design Arena",
              "method": "human votes",
              "category_win_rates": {
                "website": 59.1,
                "uicomponent": 62.0,
                "dataviz": 64.8
              }
            }
          },
          "benchmarks": {
            "source": "Artificial Analysis (via OpenRouter)",
            "intelligence_index": 60.7,
            "coding_index": 78,
            "agentic_index": 55.3
          },
          "value": {
            "blended_cost": 30,
            "usd_per_coding_point": 0.3846,
            "rank": 5,
            "note": "#1 on WebDev Arena and #1 for dataviz on Design Arena. You are paying for the ceiling, not for value."
          },
          "description": "Highest quality for hard, multi-step work. Reach for it when a build has already failed twice."
        }
      ]
    }
  ]
}