Claude Sonnet 5.5 (effort default (no --effort flag) · thinking off) on Model ladder: fix 10 real merged bugs with tests hidden

AClaude Sonnet 5.5default · thinking off · Claude Code 2.1.299 (headless -p) · cloud · 2026-10-16 · by @bobsiopener

Trust 80%Fully autonomoussuccess · first tryCompare

Estimated cost: $0.04
Wall time: 10s
Human step-ins: 0
Outcome: Pass

Compare with other models on this task →

Exact prompt (this run)

You are working in a checkout of the marshmallow-code/marshmallow repository. A user filed this bug report: --- validate.URL rejects a URL whose fragment directly follows the host ```python from marshmallow import validate validate.URL()("http://example.org#fragment") # ValidationError: Not a valid URL. validate.URL()("http://user@example.com#fragment") # same ``` `http://example.org/#fragment` and `http://example.org?a=1#fragment` are accepted, so only the bare-host + fragment case fails. RFC 3986 allows a fragment after an empty path, and links like `https://example.com#section` are common. Relatedly, with `relative=True` (and in relative-only mode) a bare same-document reference such as `#frag` is rejected, although a bare `?query` is accepted. --- Fix the bug in the library code. Dependencies are already installed (Python virtualenv in .venv; run tests with .venv/bin/pytest). Keep the change minimal and in the style of the codebase; you may add or update tests. Make sure the existing test suite still passes. You have no internet access.

Five real bug fixes merged into popular open-source libraries after the models' training cutoff (humanize, marshmallow, tenacity, nanoid, path-to-regexp). Each agent starts from the commit before the fix with only a symptom-style bug report, and the fix PR's own regression tests are applied afterwards to check the result. The size of the agent's source change is compared with the human fix. 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). Upstream fix being reproduced: https://github.com/marshmallow-code/marshmallow/pull/3016, https://github.com/marshmallow-code/marshmallow/commit/902f99c4151d1f0c4c3b8e8dfbafc7b6a76aeaa6.

Success criteria: Hidden PR tests pass; existing suite stays green; diff under 2x the human diff size.

Setup notes: Real merged bug fix: marshmallow-code/marshmallow https://github.com/marshmallow-code/marshmallow/pull/3016 (merged after the models' Jun 2026 cutoff). Agent got a history-free snapshot of the parent commit 3207be3147 with deps installed, and a symptom-only bug report written by the runner from the PR description (no fix hints). The PR's test files were overlaid only after the agent finished. Tools: Read/Write/Edit/Glob/Grep/Bash, web disabled. Diff-size flag (not scored): agent changed 2 source lines vs 2 in the human fix (allowance 20; within allowance). Claude Code's own reported cost for claude-sonnet-5.5: $0.041161. Score: 2/2. 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: $0.04
Wall time: 10s
Human step-ins: 0
Turns: 4

Model usage
Model	Input tok	Output tok	Cost
AClaude Sonnet 5.5anthropic/claude-sonnet-5.5effort default (no --effort flag) · thinking off	57,026	787	$0.04

Reproduce this run

Use the exact prompt shown above for this run (the use case page has the shared prompt).
Same setup: Claude Sonnet 5.5 (anthropic/claude-sonnet-5.5, effort default (no --effort flag) · thinking off) via Claude Code 2.1.299 (headless -p), cloud.
It counts as a success only if it meets every success criterion: Hidden PR tests pass; existing suite stays green; diff under 2x the human diff size.
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).

Run date is not in the future : passed
Model is in the dated pricing table : passed
Cost recomputed from tokens matches : passed
Cached tokens are consistent : passed
Output throughput is plausible : passed
Autonomy matches step-ins : passed
Step-ins do not exceed turns : passed
Retries agree with first-try result : passed
Evidence log SHA-256 matches : passed
Evidence log is substantial : passed

Evidence log

[
 {
  "duration_api_ms": 7775,
  "stop_reason": "end_turn",
  "session_id": "[REDACTED]",
  "total_cost_usd": 0.04116090000000001,
  "usage": {
   "input_tokens": 6,
   "cache_creation_input_tokens": 7071,
   "cache_read_input_tokens": 49949,
   "output_tokens": 787,
   "output_tokens_details": {
    "thinking_tokens": 0
   },
   "server_tool_use": {
    "web_search_requests": 0,
    "web_fetch_requests": 0
   },
   "service_tier": "standard",
   "cache_creation": {
    "ephemeral_1h_input_tokens": 7071,
    "ephemeral_5m_input_tokens": 0
   },
   "inference_geo": "global",
   "iterations": [
    {
     "input_tokens": 2,
     "output_tokens": 203,
     "cache_read_input_tokens": 19009,
     "cache_creation_input_tokens": 601,
     "cache_creation": {
      "ephemeral_5m_input_tokens": 0,
      "ephemeral_1h_input_tokens": 601
     },
     "type": "message"
    }
   ],
   "speed": "standard",
   "fallback_credit": null
  },
  "modelUsage": {
   "claude-sonnet-5-5": {
    "inputTokens": 6,
    "outputTokens": 787,
    "cacheReadInputTokens": 49949,
    "cacheCreationInputTokens": 7071,
    "webSearchRequests": 0,
    "costUSD": 0.04116090000000001,
    "contextWindow": 1000000,
    "maxOutputTokens": 128000,
    "thinkingTokens": 0,
    "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": 4,
  "subtype": "success",
  "api_error_status": null,
  "result": "`validate.URL` now accepts a fragment straight after the host, so `http://example.org#fragment` and `http://[REDACTED:email]#fragment` pass. With `relative=True`, and in relative-only mode, a bare `#frag` is also accepted now.\n\nThe fix is one character in `src/marshmallow/validate.py`. The relative-part regex allowed `/` or `?` to start a path, query or fragment, and I added `#` to that set.\n\nI ran your two examples and `#frag` in both relative modes, and all of them validate. The full suite passes (1184 tests). I didn't add new test cases for the fragment forms.",
  "ttft_ms": 2545,
  "type": "result",
  "duration_ms": 9831,
  "uuid": "[REDACTED]",
  "ttft_stream_ms": 1437,
  "time_to_request_ms": 225,
  "first_content_frame_ms": 1437,
  "queued_turn_count": 0,
  "result_index": 0
 }
]