Skip to content

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

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

Trust 80%Fully autonomoussuccess · first try
Estimated cost
$0.0089
Wall time
37s
Human step-ins
0
Outcome
Pass

Compare with other models on this task →

Summary

Exact prompt (this run)

You are working in a checkout of the ai/nanoid repository. A user filed this bug report: --- customAlphabet()/customRandom() return a constant string for a non-integer size Since 5.1.8, calling a generator with a fractional size gives broken ids on power-of-two alphabets, e.g. `customAlphabet('0123456789abcdef')(16.5)` returns `"00000000000000000"` (the first alphabet character repeated) in the browser build, and the node build is also wrong. I'd expect a fractional size to behave like its integer part (16 random characters here), in both the node and browser builds, and for `customRandom()` too. --- Fix the bug in the library code. Dependencies are already installed (node_modules). 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/ai/nanoid/pull/616, https://github.com/ai/nanoid/commit/0b2cec58ffd5ddc45978bf75a3bab46f53c76128.

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

Setup notes: Real merged bug fix: ai/nanoid https://github.com/ai/nanoid/pull/616 (merged after the models' Jun 2026 cutoff). Agent got a history-free snapshot of the parent commit ccb0e6ee55 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 8 source lines vs 6 in the human fix (allowance 20; within allowance). Claude Code's own reported cost for claude-haiku-5.5: $0.00888. 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
$0.0089
Wall time
37s
Human step-ins
0
Turns
12
Model usage
ModelInput tokOutput tokCost
AClaude Haiku 5.5anthropic/claude-haiku-5.5effort default (no --effort flag) · thinking on216,9795,946$0.0089

Reproduce this run

  1. Use the exact prompt shown above for this run (the use case page has the shared prompt).
  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: Hidden PR tests pass; existing suite stays green; diff under 2x the human diff size.
  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": 31046,
  "stop_reason": "end_turn",
  "session_id": "[REDACTED]",
  "total_cost_usd": 0.008880199999999998,
  "usage": {
   "input_tokens": 16,
   "cache_creation_input_tokens": 19663,
   "cache_read_input_tokens": 197300,
   "output_tokens": 5946,
   "output_tokens_details": {
    "thinking_tokens": 2528
   },
   "server_tool_use": {
    "web_search_requests": 0,
    "web_fetch_requests": 0
   },
   "service_tier": "standard",
   "cache_creation": {
    "ephemeral_1h_input_tokens": 19663,
    "ephemeral_5m_input_tokens": 0
   },
   "inference_geo": "global",
   "iterations": [
    {
     "input_tokens": 2,
     "output_tokens": 914,
     "cache_read_input_tokens": 31509,
     "cache_creation_input_tokens": 693,
     "cache_creation": {
      "ephemeral_5m_input_tokens": 0,
      "ephemeral_1h_input_tokens": 693
     },
     "type": "message"
    }
   ],
   "speed": "standard",
   "fallback_credit": null
  },
  "modelUsage": {
   "claude-haiku-5-5": {
    "inputTokens": 16,
    "outputTokens": 5946,
    "cacheReadInputTokens": 197300,
    "cacheCreationInputTokens": 19663,
    "webSearchRequests": 0,
    "costUSD": 0.008880199999999998,
    "contextWindow": 1000000,
    "maxOutputTokens": 128000,
    "thinkingTokens": 2528,
    "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": 12,
  "subtype": "success",
  "api_error_status": null,
  "result": "Fixed. Fractional sizes now behave like their integer part in `customRandom()` and `customAlphabet()`, in both the node and browser builds. The test suite passes (85/85, up from 83 with the two new assertions).\n\n**Cause:** the generators returned by `customRandom()` used the raw `size` in the loops. With `16.5`, the power-of-two path ran `while (j--)` from `16.5`, so it indexed `bytes[15.5]`, which is `undefined`. `undefined & mask` is `0`, so every character became `alphabet[0]`. The non-power-of-two path had the same problem with the length check: it stopped at 17 characters instead of 16. The node `customAlphabet()` looked correct only because its fast path normalizes `size |= 0` itself, while its fallback to `customRandom()` didn't.\n\n**Change:**\n- `index.js` and `index.browser.js`: in both generators of `customRandom()`, added `size |= 0` before the `!size` check, with a short comment in the same style as the existing `|=` comment in `index.js`. The browser `customAlphabet()` delegates to the same code, so it is fixed too.\n- `test/index.test.js`: added a `customRandom` test that runs for both builds and checks that `16.5` and `4.9` produce 16 and 4 characters on power-of-two and non-power-of-two alphabets, and that `customAlphabet()` returns 16 characters for `16.5`.\n\nBefore the fix, the repro gave `\"[REDACTED:credit_card]\"` in the browser and `\"cdacd\"` for a non-power-of-two `4.5` in node. After it, both return the correct length with random characters.\n\n`oxlint` could not run here: the installed version fails to load its config on the available Node 20.19.2. This is a toolchain issue, not something this change introduced. I did not run the formatter check separately, so treat formatting as unverified.",
  "ttft_ms": 3497,
  "type": "result",
  "duration_ms": 36765,
  "uuid": "[REDACTED]",
  "ttft_stream_ms": 3101,
  "time_to_request_ms": 205,
  "first_content_frame_ms": 3102,
  "queued_turn_count": 0,
  "result_index": 0
 }
]

Get new runs and ranking changes weekly