Skip to content

Claude Haiku 5.5 (effort default (no --effort flag) · thinking on) on Boat fuel range with a reserve over a planned trip

AClaude Haiku 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
$0.0056
Wall time
29s
Human step-ins
0
Outcome
Pass

Compare with other models on this task →

Summary

Exact prompt

# Boat fuel range with a reserve over a planned trip `input.json` (synthetic; not for real trip planning): `tank_l`, `fuel_l` on board, `reserve_pct` (of the tank), `burn_table` (rows `rpm`, `speed_kn` through the water, `lph` litres per hour; not sorted), and `legs` (`distance_nm`, `rpm`, `current_kn`: positive = with the boat, negative = against). Write `solve.py` (Python 3 standard library only); `python solve.py input.json answer.json`. Fresh inputs are used too. ## Rules 1. Speed and burn at an rpm: sort the table by rpm, interpolate each linearly, clamp to the first/last row outside the range. 2. Each leg in order: SOG = speed + current_kn. If SOG <= 0 the leg cannot be run: append `{"sog_kn", "hours": null, "fuel_l": null, "remaining_l"}` and stop processing legs (this leg is the dry leg if none yet, with `dry_after_nm_into_leg` 0 if none yet). Otherwise hours = distance / SOG, fuel = lph * hours, remaining -= fuel (remaining may go negative; keep going). 3. `reserve_l` = tank * pct / 100. `reserve_breach_leg` = index (0-based) of the first leg after which remaining < reserve, or null. `runs_dry_leg` = first leg whose fuel exceeds the fuel left at its start, and `dry_after_nm_into_leg` = fuel left at its start / lph * SOG; both null if never. 4. Economy: the table row with the greatest speed_kn / lph (compared at 6 decimals; ties -> lower rpm). `economy_range_nm` = max(0, fuel_l - reserve) * that ratio (still water). Round speeds and hours to 3, litres and nm to 2, `economy_nm_per_l` to 4. ## answer.json `{"reserve_l", "legs": [{"sog_kn", "hours", "fuel_l", "remaining_l"}, ...], "reserve_breach_leg", "runs_dry_leg", "dry_after_nm_into_leg", "economy_rpm", "economy_nm_per_l", "economy_range_nm"}` ## Scoring Hidden checks (partial credit; all must pass): `leg_speeds_and_burn` (first three legs), `remaining_fuel`, `reserve_and_dry`, `economy`, `fresh_run_1`, `fresh_run_2`. Tolerance 0.011 (0.0002 for nm per litre). 60-second limit per run.

Unsorted burn table with clamped interpolation, current added to speed (including a leg where the boat cannot make way), separate reserve-breach and run-dry legs, and the economy-rpm tie rule. 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. leg_speeds_and_burn 2. remaining_fuel 3. reserve_and_dry 4. economy 5. fresh_run_1 6. fresh_run_2

Setup notes: Use Case Scout batch 10 case k11-fuel-range-reserve (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. Claude Code's own reported cost for claude-haiku-5.5: $0.005565. Score: 6/6. 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.

Metrics

Estimated cost
$0.0056
Wall time
29s
Human step-ins
0
Turns
5
Model usage
ModelInput tokOutput tokCost
AClaude Haiku 5.5anthropic/claude-haiku-5.5effort default (no --effort flag) · thinking on110,9044,136$0.0056

Reproduce this run

  1. Use the use case prompt on the use case page.
  2. Same setup: Claude Haiku 5.5 (anthropic/claude-haiku-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. leg_speeds_and_burn 2. remaining_fuel 3. reserve_and_dry 4. economy 5. fresh_run_1 6. 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": 27560,
  "stop_reason": "end_turn",
  "session_id": "[REDACTED]",
  "total_cost_usd": 0.005565100000000001,
  "usage": {
   "input_tokens": 10,
   "cache_creation_input_tokens": 12564,
   "cache_read_input_tokens": 98330,
   "output_tokens": 4136,
   "output_tokens_details": {
    "thinking_tokens": 2022
   },
   "server_tool_use": {
    "web_search_requests": 0,
    "web_fetch_requests": 0
   },
   "service_tier": "standard",
   "cache_creation": {
    "ephemeral_1h_input_tokens": 12564,
    "ephemeral_5m_input_tokens": 0
   },
   "inference_geo": "global",
   "iterations": [
    {
     "input_tokens": 2,
     "output_tokens": 916,
     "cache_read_input_tokens": 24337,
     "cache_creation_input_tokens": 766,
     "cache_creation": {
      "ephemeral_5m_input_tokens": 0,
      "ephemeral_1h_input_tokens": 766
     },
     "type": "message"
    }
   ],
   "speed": "standard",
   "fallback_credit": null
  },
  "modelUsage": {
   "claude-haiku-5-5": {
    "inputTokens": 10,
    "outputTokens": 4136,
    "cacheReadInputTokens": 98330,
    "cacheCreationInputTokens": 12564,
    "webSearchRequests": 0,
    "costUSD": 0.005565100000000001,
    "contextWindow": 1000000,
    "maxOutputTokens": 128000,
    "thinkingTokens": 2022,
    "canonicalModel": "claude-haiku-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": 5,
  "subtype": "success",
  "api_error_status": null,
  "result": "I wrote `solve.py` in the work directory and ran it on `input.json`, which produced `answer.json`. It uses only the standard library and finishes well within the 60-second limit.\n\nKey results: reserve is 60.0 L, the reserve is first breached after leg 4, and the fuel runs dry on leg 5 about 9.31 nm into that leg. The most economical row is 1400 rpm at 0.6429 nm/L, giving 124.71 nm of range above reserve.\n\nI spot-checked leg 0 by hand (rpm 2100 interpolated between the 1800 and 2200 rows, current \u22120.9 kn) and the economy and dry-leg figures against the rules. I have not run the hidden checks, and I did not test fresh inputs beyond reading the code.\n\nInterpretation choices:\n- Intermediate values use full precision; only the written output is rounded.\n- The \"dry leg\" with `sog <= 0` sets `runs_dry_leg` to that leg with `dry_after_nm_into_leg` 0 only if no earlier leg has already run dry.\n- `reserve_breach_leg` and `runs_dry_leg` record only the first occurrence.\n- Duplicate rpm rows are handled without dividing by zero, but I did not define a specific choice between them.\n\nThe directory is not a git repository, so I did not commit anything.",
  "ttft_ms": 8926,
  "type": "result",
  "duration_ms": 27956,
  "uuid": "[REDACTED]",
  "ttft_stream_ms": 6695,
  "time_to_request_ms": 235,
  "first_content_frame_ms": 6695,
  "queued_turn_count": 0,
  "result_index": 0
 }
]

Get new runs and ranking changes weekly