Benchmark Indonesian

Indonesian spelling correction

Accepting or rejecting spell-checker suggestions on real text.Read more
Part of Benchmark Nusantara

Our own collection of benchmarks for Indonesian and its regional languages. Each set stands on its own, shares the same method and guards, and its data is published in one repository.

54 itemsLast tested
GitHub

12

Models tested

54/ 16

Questions / held back

12

Runs recorded

Quick summary

Highest score

76.3%

29 of 38 answered correctly

Gemma 4 31BGoogle
The gap to second place is 4 items, and this page makes no claim about the order between models whose run-to-run ranges overlap.

Cheapest

$0.000013

per correct answer

Solar Pro 4Upstage
Computed only among models scoring at least half of the top score on this board, because a model that gets a lot wrong always looks cheapest. This model scored 60.5%.

Lowest latency

556 ms

median wait

Mistral Small 3Mistral
It scored 57.9%, so fast here does not mean accurate.

Results

How many questions each model answered correctly, from the exact same set. The bars show the count of correct answers, not a general ranking of model ability.

open answer

The model writes its own answer, with no options offered. Some are scored automatically against a reference, some are read one by one by a native speaker using a weighted rubric.

temperature 0token cap variesKeterangan
12 models · 38 items scored

A model whose name does not end in a date is called through an id that does not pin the version. The provider may swap the model behind that id at any time without notice, so the figures apply to whichever version was active when the run happened.

How these figures are computed
phrasing open-py1Keteranganscoring open-py1Keteranganquestion version 27af2c184856Keterangananswer key version 82bbf3cc2f8aKeterangan

Per-model detail

The same figures as the chart above, plus the cost per correct answer, the wait time, and the date of the last run. Any column that cannot be guessed from its name carries its own explanation.

open answer
Results per model on the open answer task: correct answers, cost per correct answer, wait time, and the date of the last run.
ModelCorrectKeteranganHeld backKeterangan$/correctKeteranganLatencyKeteranganLast runKeterangan
Gemma 4 31BGoogle29/3876.3%13/16$0.000031$0.00131,021 msmedian
GPT-5.6 LunaOpenAI25/3865.8%12/16$0.000141$0.00522,701 msmedian
Ling-3.0-flashAnt Group25/3865.8%12/16$0.000042$0.00161,410 msmedian
Gemma-SEA-LION v4.5 E2B-ITlocal25/3865.8%11/16-836 msmedian
DeepSeek V4 Flash 0731DeepSeek23/3860.5%12/16$0.000112$0.00395,016 msmedian
Solar Pro 4Upstage23/3860.5%10/16$0.000013$0.000433,811 msmedian
Mistral Small 3Mistral22/3857.9%9/16$0.000019$0.0006556 msmedian
Qwen3 30B A3B Instruct 2507Alibaba21/3855.3%13/16$0.000030$0.0010596 msmedian
Llama3 8B CPT Sahabat-AI v1 Instructlocal21/3855.3%11/16-301 msmedian
Hunyuan A13B InstructTencent21/3855.3%7/16$0.000068$0.00191,474 msmedian
Claude Haiku 4.5Anthropic20/3852.6%12/16$0.000508$0.01631,266 msmedian
Llama 3.1 8B Instructlocal19/3850.0%11/16-443 msmedian

Models marked local run on our own hardware, so there is no bill to record and the wait time measures our machine, not a service anyone else can buy. The score is still comparable: the items, the key, and the temperature are identical to every other row here.

Every figure in this table can be recomputed from the raw data. Open the data

Cost and tokensCheapest per correct answer: Solar Pro 4 ($0.000013). Most expensive: Claude Haiku 4.5 ($0.000508).

All cost figures are in US dollars.

Cost per correct answer

The model with the highest score is not always the cheapest one. This divides the cost of one full test by its number of correct answers.

Cost per correct answer

Cost of one full run

The raw figure, before dividing by the number of correct answers. Every model answered the exact same questions, so this is directly comparable: it is what you pay to run this benchmark once on each model.

Cost of one full run

Tokens used

Input and output are kept apart because they are priced differently, often tenfold. A long output bar means the model talks a lot, and that is where almost all of the cost difference between models comes from. A model with no bar has no token record, which is not the same as zero.

inputoutput
Ling-3.0-flash13,118.0 · 16,337.0
DeepSeek V4 Flash 073111,085.0 · 13,176.0
GPT-5.6 Luna10,655.0 · 6,923.0
Claude Haiku 4.514,517.0 · 349.0
Hunyuan A13B Instruct12,566.0 · 260.0
Llama 3.1 8B Instruct12,831.0 · 255.0
Solar Pro 413,819.0 · 226.0
Mistral Small 311,546.0 · 221.0
Llama3 8B CPT Sahabat-AI v1 Instruct12,831.0 · 205.0
Gemma 4 31B10,323.0 · 190.0
Qwen3 30B A3B Instruct 250712,782.0 · 186.0
Gemma-SEA-LION v4.5 E2B-IT9,909.0 · 184.0
Tokens used

Benchmark