{
  "schema": "genie-recipe.v1",
  "id": "sovereign-coding-room",
  "name": "Sovereign Coding Room",
  "stage": "benchmarked",
  "commercial": false,
  "verified_at": "2026-08-02T21:25:00Z",
  "model": {
    "id": "deepseek-ai/DeepSeek-V4-Flash-0731",
    "checkpoint_gib": 155.43,
    "parameters_billion": 304.18,
    "license": "MIT"
  },
  "runtime": {
    "engine": "vLLM 0.25.0",
    "speculative_decoder": "DSpark",
    "speculative_tokens": 7,
    "context_tokens_tested": 8192,
    "max_sequences": 8,
    "gpu": "1x NVIDIA B300 288 GB",
    "gpu_hour_usd_measured": 7.39
  },
  "evidence": {
    "generic_output_tok_s": 16.17,
    "coding_median_output_tok_s": 32.17,
    "batch_8_aggregate_output_tok_s": 169.74,
    "batch_8_median_latency_s": 6.03,
    "batch_16_aggregate_output_tok_s": 104.24,
    "coding_first_pass": "4/5",
    "traceback_repair": "0/1",
    "responses_api": "HTTP 200, exact READY",
    "raw_cold_start_minutes": [21.53, 29.30],
    "same_volume_restart_minutes": [4.47, 4.58]
  },
  "modeled_cost": {
    "best_compute_per_million_output_tokens_usd": 12.09,
    "hosted_output_per_million_tokens_usd": 0.28,
    "sovereign_premium_x": 43.2,
    "always_on_monthly_usd": 5394.70
  },
  "required_policy": {
    "minimum_queued_tasks": 4,
    "concurrency": 8,
    "compile_and_test": true,
    "maximum_repair_attempts": 1,
    "idle_teardown_minutes": 5
  },
  "known_boundaries": [
    "The advertised 1M-token context and 384K maximum output were not tested.",
    "The five-task coding suite is a smoke test, not HumanEval or SWE-bench.",
    "The package is not wired to Genie's payment and lifecycle controller yet.",
    "No persistent storage is provisioned; current provider burn is zero."
  ],
  "files": {
    "launch": "/recipes/deepseek-v4-flash-0731/serve.sh",
    "documentation": "/recipes/deepseek-v4-flash-0731/README.md"
  }
}