/ torvalds / linux
Updated 7/21/2026
Linux kernel source tree
Key Takeaways
- GPT-5.6 Sol has the clearest lead in the collection, resolving 63 of 85 tasks—five more than the next-best models.
- GPT-5.6 Terra is the strongest efficiency option at 56 of 85 tasks, $0.24 per test, 102 seconds, and 0.23M tokens per test.
- Claude Fable 5 and Muse Spark 1.1 resolve 69 of 85 tasks together; Kimi K3 also uniquely solves two tasks despite its middle-of-table score.
Cost / Test vs. Accuracy
Grok 4.5 is cheapest at $0.17 per test with 55 tasks resolved, while GPT-5.6 Terra adds one task for $0.24 and the leader reaches 63 for $1.09.
Latency vs. Accuracy
GPT-5.6 Terra resolves 56 tasks in 102 seconds and Grok 4.5 resolves 55 in 109 seconds; GPT-5.6 Sol takes 309 seconds for seven additional tasks.
Average token use / test
Tasks with failures
| Model | 32a92f8 | e55d98e | 1954c4f | 23e6a1c | 189f164 | ec69c9e | 2687c84 | e19e1b4 | 52f657e | aa54b1d | 5401b9a | 551d442 | 7dff99b | f20d61c | 805d5a2 | 1f5ffc6 | ebb0f38 | 53676e4 | c1fa0bb | b29a7a8 | 9197e59 | bfbc0b5 | 31e62c2 | 7893cc1 | 284922f | 614da1d | 8c2e52e | 983d014 | 00a7d39 | 87c9e88 | b69053d | 99b773d | fea4e31 | ad8cccc | 5bebe8d | f90fff1 | 4b4bd8c | 318e8c3 | af7809f | 488ef35 | 2e4b28c | d440148 | 12eef14 | 3af870a | 9f35e33 | 82efd56 | d2ea4d2 | 9d7a057 | 323bbfc | 05e8d26 | 2985dae | 4cc5373 | d0ca0df | 7881cd6 | a48f822 | 478ad02 | f2c61db | 1e83ccd |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| GPT-5.6 Sol | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Claude Opus 4.8 | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Claude Fable 5 | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Muse Spark 1.1 | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| GPT-5.6 Terra | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Claude Opus 4.7 | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Grok 4.5 | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| GPT 5.5 | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Claude Sonnet 5 | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| GPT-5.6 Luna | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Kimi K3 | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Claude Haiku 4.5 (Nonthinking) | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Gemini 3.1 Pro Preview (02/26) | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Gemini 3.5 Flash | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| GLM 5.2 (Fireworks) |
Task detail
32a92f8c8932Issue statement
The typed kernel allocation helpers (kmalloc_obj, kzalloc_obj, their plural forms, the vmalloc variants, and flexible-structure variants) already default to GFP_KERNEL when no allocation flag is supplied. Audit the straightforward call sites that explicitly pass GFP_KERNEL as their final argument and convert them to the default form by omitting that redundant argument. This includes simple calls split across lines throughout architecture code and core subsystems. Calls using a non-default allocation policy must retain their explicit flag, and allocation behavior must otherwise remain unchanged.
Hidden tests
diff --git a/tools/testing/valsmith/test_alloc_obj_defaults.py b/tools/testing/valsmith/test_alloc_obj_defaults.pynew file mode 100644index 000000000..f4b0a2b18--- /dev/null+++ b/tools/testing/valsmith/test_alloc_obj_defaults.py@@ -0,0 +1,64 @@+#!/usr/bin/env python3+import argparse+import sys+import xml.etree.ElementTree as ET+from pathlib import Path+++CASES = {+ "arch/arm64/kernel/vdso.c":+ "kzalloc_objs(struct page *, vdso_info[abi].vdso_pages)",+ "arch/powerpc/kvm/book3s_pr.c":+ "kzalloc_obj(*vcpu->arch.shadow_vcpu)",+ "arch/sparc/kernel/smp_64.c":+ "kzalloc_flex(*hdesc, maps, num_kernel_image_mappings)",+ "block/blk-crypto-profile.c":+ "kvmalloc_objs(profile->slot_hashtable[0], slot_hashtable_size)",+ "drivers/acpi/scan.c":+ "kzalloc_objs(*r, ret + 1)",+ "drivers/gpu/drm/drm_atomic.c":+ "kzalloc_obj(*state)",+ "net/unix/af_unix.c":+ "kvmalloc_objs(spinlock_t, UNIX_HASH_SIZE)",+ "sound/usb/mixer.c":+ "kzalloc_obj(*cval)",+}+++def normalized(path):+ return " ".join(Path(path).read_text(encoding="utf-8").split())+++def check_default_kernel_allocations():+ missing = []+ for path, call in CASES.items():+ if call not in normalized(path):+ missing.append(f"{path}: expected {call}")+ if missing:+ raise AssertionError("Default GFP_KERNEL allocation form missing:\n" + "\n".join(missing))+++def main():+ parser = argparse.ArgumentParser()+ parser.add_argument("--junitxml", required=True)+ args = parser.parse_args()+ suite = ET.Element("testsuite", name="alloc_obj_defaults", tests="1")+ case = ET.SubElement(suite, "testcase", classname="alloc_obj_defaults",+ name="default_kernel_allocations_omit_redundant_flag")+ failed = False+ try:+ check_default_kernel_allocations()+ except Exception as error:+ failed = True+ suite.set("failures", "1")+ ET.SubElement(case, "failure", message=str(error)).text = str(error)+ else:+ suite.set("failures", "0")+ report = Path(args.junitxml)+ report.parent.mkdir(parents=True, exist_ok=True)+ ET.ElementTree(suite).write(report, encoding="utf-8", xml_declaration=True)+ return 1 if failed else 0+++if __name__ == "__main__":+ sys.exit(main())
Head-to-head
Accuracy
74.12%Δ 5.88%
68.24%
Cost / test
$1.09Δ $0.33
$1.43
Latency
309sΔ 120s
428s
Cost distribution
Latency distribution
Input token distribution
Output token distribution
Task outcomes
85 tasks