Skip to content

Claude Sonnet 5.5 (effort default (no --effort flag) · thinking on) on Go/no-go departure time from a gridded wind file

AClaude Sonnet 5.5default · thinking on · Claude Code 2.1.299 (headless -p) · cloud · 2026-10-20 · by @bobsiopener

Trust 80%Fully autonomoussuccess · first try
Estimated cost
$0.06
Wall time
21s
Human step-ins
0
Outcome
Pass

Compare with other models on this task →

Summary

Exact prompt

# Go/no-go departure time from a gridded wind file `input.json` (synthetic; not a forecast) names `grid.json` and gives `from` and `to` [lat, lon], `passage_hours` D (integer), `latest_departure_h`, and `limits`. `grid.json` has `lats`, `lons`, `times_h` (ascending) and fields `u_kt`, `v_kt` (eastward/northward wind components, knots, the direction the air moves TOWARD) and `gust_kt`, each indexed `[time][lat][lon]`. Write `solve.py` (Python 3 standard library only); `python solve.py input.json answer.json` writes `answer.json` and `window.svg`. Fresh inputs are used too. ## Rules 1. Interpolation of each field: bilinear in (lat, lon) at the two bracketing grid times, then linear in time. Outside a grid axis, clamp to its edge. 2. Course = atan2((lon_to - lon_from) * cos(mean lat), lat_to - lat_from) in degrees mod 360. 3. For every integer departure hour t0 = 0 .. latest_departure_h: sample k = 0..D at time t0 + k and position from + (to - from) * k / D (straight in lat/lon). At each sample: wind speed = hypot(u, v); wind FROM direction = atan2(-u, -v) mod 360; off = smallest angle between it and the course. The sample fails, checked in this order, with reason `wind@<t>` if speed > max_wind_kt, `gust@<t>` if gust > max_gust_kt, `headwind@<t>` if off < min_wind_angle_deg and speed > headwind_max_kt. Stop at the first failing sample; that departure is not ok and its reason is recorded (t as an integer). Otherwise ok with reason null. 4. `earliest_ok_h`: first ok departure or null. `longest_ok_run`: [first, last] departure hours of the longest run of consecutive ok departures (earliest wins ties), or null. ## Visual output: window.svg Static, self-contained SVG (root `<svg>`, no scripts, images, links or external assets; renders headless) with an element `id="timeline"` (one mark per departure hour) and an element `id="best-departure"`. The drawing itself is not graded. ## answer.json `{"course_deg", "departures": [{"depart_h", "ok", "reason"}, ...], "earliest_ok_h", "longest_ok_run"}` ## Scoring Hidden checks (partial credit; all must pass): `course` (within 0.11), `departure_flags` (ok flags), `failure_reasons` (exact), `best_departure`, `window_svg_static` (window.svg exists, follows the static rules, has the ids above), `fresh_run_1`, `fresh_run_2`. 60-second limit per run.

Space-time interpolation of vector components (not speed and direction), the toward/from convention, ordered failure reasons per sample along a moving position, and run-length selection with ties. Original benchmark task written for usecaserank.com (not based on a specific public post). Headless, offline run on a reproducible fixture. Each run's notes say how its cost was measured (estimates are marked).

Benchmark code for comparing AI models, not navigation or safety software.

Success criteria: 1. course 2. departure_flags 3. failure_reasons 4. best_departure 5. window_svg_static 6. fresh_run_1 7. fresh_run_2

Setup notes: Use Case Scout batch 10 case k10-wind-grid-departure (offline fixture). Prepared with batch10/prepare.sh; hidden key/tests kept outside the agent's work dir. Prompt = TASK.md verbatim. Scored with the scout's scorer: `python score.py WORKDIR KEYDIR`. Tools: Read/Write/Edit/Glob/Grep/Bash, web disabled. Single run, no retries. Visual output (window.svg) rendered offline to PNG with google-chrome headless, network blocked, 1280x800 at ~3 s after load. Claude Code's own reported cost for claude-sonnet-5.5: $0.064027. Score: 7/7. Cost is an ESTIMATE: measured tokens x pricing.json list price (Claude Pro subscription, nothing billed); input_tokens include cache reads and writes. Single run.

Estimated cost List-price estimate; not actually billed.

Output

Metrics

Estimated cost
$0.06
Wall time
21s
Human step-ins
0
Turns
3
Model usage
ModelInput tokOutput tokCost
AClaude Sonnet 5.5anthropic/claude-sonnet-5.5effort default (no --effort flag) · thinking on59,9342,284$0.06

Reproduce this run

  1. Use the use case prompt on the use case page.
  2. Same setup: Claude Sonnet 5.5 (anthropic/claude-sonnet-5.5, effort default (no --effort flag) · thinking on) via Claude Code 2.1.299 (headless -p), cloud.
  3. It counts as a success only if it meets every success criterion: 1. course 2. departure_flags 3. failure_reasons 4. best_departure 5. window_svg_static 6. fresh_run_1 7. fresh_run_2
  4. Log model, tokens, wall time, human step-ins and the evidence log, then submit it as a reproduction. A different model? Submit a head-to-head rerun.

How runs are scored: Methodology.

Was this useful?

⇄ Rerun on another model

Automated checks

Trust 80% comes from these checks (how Trust % works).

Evidence log
[
 {
  "duration_api_ms": 19818,
  "stop_reason": "end_turn",
  "session_id": "[REDACTED]",
  "total_cost_usd": 0.0640267,
  "usage": {
   "input_tokens": 6,
   "cache_creation_input_tokens": 9021,
   "cache_read_input_tokens": 50907,
   "output_tokens": 2284,
   "output_tokens_details": {
    "thinking_tokens": 85
   },
   "server_tool_use": {
    "web_search_requests": 0,
    "web_fetch_requests": 0
   },
   "service_tier": "standard",
   "cache_creation": {
    "ephemeral_1h_input_tokens": 9021,
    "ephemeral_5m_input_tokens": 0
   },
   "inference_geo": "global",
   "iterations": [
    {
     "input_tokens": 2,
     "output_tokens": 328,
     "cache_read_input_tokens": 19570,
     "cache_creation_input_tokens": 1990,
     "cache_creation": {
      "ephemeral_5m_input_tokens": 0,
      "ephemeral_1h_input_tokens": 1990
     },
     "type": "message"
    }
   ],
   "speed": "standard",
   "fallback_credit": null
  },
  "modelUsage": {
   "claude-sonnet-5-5": {
    "inputTokens": 6,
    "outputTokens": 2284,
    "cacheReadInputTokens": 50907,
    "cacheCreationInputTokens": 9021,
    "webSearchRequests": 0,
    "costUSD": 0.0640267,
    "contextWindow": 1000000,
    "maxOutputTokens": 128000,
    "thinkingTokens": 85,
    "canonicalModel": "claude-sonnet-5-5",
    "provider": "firstParty",
    "costBasis": "list"
   }
  },
  "permission_denials": [],
  "terminal_reason": "completed",
  "fast_mode_state": "off",
  "fast_mode_disabled_reason": "sdk_opt_in_required",
  "subagent_stats": {
   "spawned": 0,
   "requested": {
    "background": 0,
    "foreground": 0,
    "unset": 0
   },
   "started_in_background": 0,
   "max_depth": 0,
   "spawned_by_subagents": 0,
   "completed": 0,
   "failed": 0,
   "killed": {
    "parent": 0,
    "user": 0,
    "system": 0
   },
   "refused": {
    "depth_limit": 0,
    "concurrency_limit": 0,
    "budget": 0
   },
   "by_type": {}
  },
  "safety_stops": 0,
  "is_error": false,
  "num_turns": 3,
  "subtype": "success",
  "api_error_status": null,
  "result": "I wrote `solve.py` and ran it on the supplied `input.json`. It wrote `answer.json` and `window.svg` without errors. I haven't checked the output against any reference, and the hidden checks and fresh inputs haven't been run.\n\nOn this input:\n- **Course:** about 39.98\u00b0.\n- **Earliest OK departure:** hour 12.\n- **Longest OK run:** hours 12 to 30. That is every departure from 12 through `latest_departure_h`.\n- **Early departures:** hours 0 and 1 fail on `wind@6`, hours 2 to 6 on `gust@6`, and hours 7 to 11 fail on `wind` or `gust` at later samples.\n\nSample times past the end of the grid (48 h here) clamp to the last time step, as rule 1 says. `window.svg` has one coloured mark per departure hour in `#timeline` and an outline around the longest OK run as `#best-departure`.",
  "ttft_ms": 2666,
  "type": "result",
  "duration_ms": 20183,
  "uuid": "[REDACTED]",
  "ttft_stream_ms": 2131,
  "time_to_request_ms": 228,
  "first_content_frame_ms": 2131,
  "queued_turn_count": 0,
  "result_index": 0
 }
]

Get new runs and ranking changes weekly