Benchmark Almanac

LocalSearchBench

DeepSeek-V3.2 35.6% leads since December 2025
Not in model cards

Agents search merchant listings to answer 900 questions about local services across nine Chinese cities. A model judges each answer against a reference.

The authors publish scores for 11 models in 16 configurations. DeepSeek V3.2 with thinking enabled scored 35.60% on answer correctness.

Released
December 2025
Kind
Questions that need search
Level
Everyday
Scored
Graded by a model
Answers
Agent task
Tasks
All public
Measures
Capability
Language
Chinese
In and out
Text → Text, Actions
Interaction
Agentic
Environment
APIs and tools
Items from
Real work and use
Refresh
One fixed set
Run by
Anyone
In views
Agents
Strongest model tested
o3 #84 on the capability index
Last model added
December 2025
Scores refreshed
November 20th, 2026
11
Models scored on its own leaderboard

DeepSeek-V3.2 leads with 35.6% among models released by December 2025

Best score per model from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

The newest model with a score here came out in December 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 DeepSeek-V3.2 w/ thinking 35.6%
  2. 2 GLM-4.6 w/ thinking 32.8%
  3. 3 Gemini 2.5 Pro Jun 2025 32.3%
  4. 4 o3 high 31.5%
  5. 5 LongCat-Flash-Thinking 30.7%
  6. 6 Qwen3-235B-A22B w/ thinking 30.2%
  7. 7 Qwen3-32B w/ thinking 25.6%
  8. 8 LongCat-Flash-Chat 25.3%
  9. 9 Qwen3-14B w/ thinking 25.2%
  10. 10 Gemini 2.5 Flash Sep 2025 21%
  11. 11 GPT-4.1 18.5%

Source: Official LocalSearchBench leaderboard

#ModelScoreSettingDateSource
1 DeepSeek-V3.2 35.6% w/ thinking Dec 2025 Official leaderboard
2 GLM-4.6 32.8% w/ thinking Sep 2025 Official leaderboard
3 Gemini 2.5 Pro Jun 2025 32.3% Jun 2025 Official leaderboard
4 o3 31.5% high Apr 2025 Official leaderboard
5 LongCat-Flash-Thinking 30.7% Sep 2025 Official leaderboard
6 Qwen3-235B-A22B 30.2% w/ thinking Apr 2025 Official leaderboard
7 Qwen3-32B 25.6% w/ thinking Apr 2025 Official leaderboard
8 LongCat-Flash-Chat 25.3% Aug 2025 Official leaderboard
9 Qwen3-14B 25.2% w/ thinking Apr 2025 Official leaderboard
10 Gemini 2.5 Flash Sep 2025 21% Sep 2025 Official leaderboard
11 GPT-4.1 18.5% Apr 2025 Official leaderboard

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
DeepSeek-V3.2
w/o thinking w/ thinking
35.6%
GLM-4.6
w/o thinking w/ thinking
32.8%
Qwen3-235B-A22B
w/o thinking w/ thinking
30.2%
Qwen3-32B
w/o thinking w/ thinking
25.6%
Qwen3-14B
w/o thinking
25.2%

Source: Official LocalSearchBench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
DeepSeek-V3.2 w/ thinking 35.6%
GLM-4.6 w/ thinking 32.8%
DeepSeek-V3.2 w/o thinking 32.7%
Gemini 2.5 Pro Jun 2025 32.3%
o3 high 31.5%
LongCat-Flash-Thinking 30.7%
Qwen3-235B-A22B w/ thinking 30.2%
GLM-4.6 w/o thinking 29%
Qwen3-32B w/ thinking 25.6%
LongCat-Flash-Chat 25.3%
Qwen3-14B w/ thinking 25.2%
Qwen3-14B w/o thinking 24.2%
Qwen3-235B-A22B w/o thinking 21.2%
Gemini 2.5 Flash Sep 2025 21%
Qwen3-32B w/o thinking 19.8%
GPT-4.1 18.5%

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

DeepSeek
Own leaderboard Not tested since Dec 2025

Source: Official LocalSearchBench leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) DeepSeek-V3.2 Gemini 2.5 Pro GPT-4.1 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) DeepSeek-V3.2 o3

Source: Official LocalSearchBench leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Knowledge

All 332 in Knowledge
  • MMLU-Pro In use Harder 10-option version of MMLU with reasoning-focused questions across 14 subjects. Intern-S2-397B 89.8%
  • MMLU In use 57-subject multiple-choice exam questions from elementary to professional level. Scored by accuracy. Seed1.8 92.3%
  • SuperGPQA In use 26,529 graduate-level multiple-choice questions across 285 disciplines. Accuracy. Qwen 3.6 Max (Preview) 73.9%
  • WideSearch In use 200 bilingual broad info-seeking tasks: agent collects many atomic facts into a table. Scored by success rate and row/item F1. Tencent Hy4 preview 83.9%
  • BBEH micro average In use Models solve harder replacements for each of the 23 BIG-Bench Hard tasks, 4,520 examples in total. The score is the share of all examples answered correctly. Gemma 4 31B IT 74.4%
  • HLE-Diamond In use Models answer 1,000 cleaned expert questions from Humanity's Last Exam, 500 on reasoning and 500 on knowledge, closed book. Graded on exact answers. Claude Fable 5.2 77.9%

Data audit

9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 11 months ago, in December 2025, for DeepSeek-V3.2. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (11 entries).
  • Pass Models scored 11 models we track.
  • Warning Newest score The newest came 11 months ago, in December 2025, for DeepSeek-V3.2.
  • Note Frontier models None of the top 10 has a score. The strongest tested is o3 (#84).
  • Pass Headroom The best is 35.6%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on localsearchbench.github.io: 16 rows for models we track at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check