Academic

IOI

Updated 10/29/2026

International Olympiad in Informatics

As of October 29, 2026, Gemini 4 Argon and GPT-6 Astra tie for first on IOI with 100.00%, followed by GPT-6.1 Sol (96.89%).

IOI v2Informatics olympiad programming problems
IOI v2•Oct 29, 2026
ACCURACY
Lab:
Pinch or Ctrl/Cmd + scroll to zoom. Once zoomed, scroll or swipe with two fingers to pan, or drag to pan. Shift + drag draws a box to zoom into; turn on box zoom to draw one without Shift. Plus and minus keys zoom, arrow keys pan, 0 resets.

IOI leaderboard

Rank Model Accuracy Cost / Test Tokens / Test Duration / Test
1 Gemini 4 Argon 100.00% $5.31 8.22m 24m19s
2 GPT-6 Astra 100.00% $6.50 963.67k 31m40s
3 GPT-6.1 Sol 96.89% $1.26 967.87k 31m31s
4 Claude Opus 5.5 95.06% $4.35 4.70m 17m15s
5 GPT-5.6 Sol 91.17% $7.98 4.50m 1h02m
6 Claude Fable 5.1 90.78% $11.20 6.28m 30m06s
7 GPT-5.6 Terra 87.61% $8.66 11.51m 1h23m
8 Claude Opus 5 84.33% $16.48 16.24m 56m42s
9 Claude Sonnet 5.5 83.06% $5.81 18.05m 45m51s
10 GPT-6 Sol 82.61% $2.70 2.78m 30m38s
11 Kimi K3.1 Pro 81.89% $5.02 9.52m 49m25s
12 Wonder Full 79.83% $7.08 7.00m 1h31m
13 Qwen 3.8 Max 68.89% $9.47 28.08m 1h47m
14 GLM 5.3 68.44% $7.67 20.34m 1h46m
15 Gemini 3.7 Flash 67.83% $3.40 7.74m 21m09s
16 Wonder 66.94% $4.54 7.79m 25m02s
17 GPT-5.6 Luna 61.78% $0.58 10.37m 1h10m
18 Hy4 Preview 59.33% $1.35 11.66m 1h01m
19 Grok 4.7 57.72% $12.71 10.87m 46m48s
20 Gemini 3.8 Flash 56.94% $3.98 13.61m 14m20s
21 Muse Spark 1.3 Max 56.56% $1.73 4.33m 28m58s
22 GPT-6 Luna 55.56% $0.20 4.46m 35m47s
23 GPT 5.3 Codex 53.83% $2.68 4.42m 29m52s
24 GLM 5.3 Flash 52.50% $0.43 18.18m 1h32m
25 Gemini 3.1 Pro Preview (02/26) 51.83% $1.25 1.28m 12m03s
26 DeepSeek V4 Pro 0813 51.61% $2.15 21.44m 1h07m
27 Kimi K3 48.94% $14.17 21.86m 4h18m
28 MiMo V2.6 Flash 47.67% $0.22 20.85m 1h49m
29 Grok 4.6 47.61% $8.18 5.53m 3h57m
30 Claude Haiku 5.5 47.28% $0.83 5.81m 12m34s
31 Ember-1 46.33% $24.46 56.12m 1h47m
32 Mistral Large 4 Preview 45.28% $2.73 12.58m 1h24m
33 Claude Sonnet 5 45.00% $12.89 38.28m 59m44s
34 Muse Spark 1.3 43.94% $2.36 5.07m 29m51s
35 Grok 4.5 40.56% $5.64 6.47m 1h27m
36 DeepSeek V4.1 Flash 40.28% $0.24 6.71m 18m30s
37 MiMo V2.6 Pro 39.33% $0.84 7.10m 2h26m
38 Qwen 3.8 27B 39.06% $3.55 17.41m 1h39m
39 Step 5 Preview 38.06% $2.03 14.93m 1h20m
40 Gemini 3.6 Flash 35.06% $8.67 39.91m 48m23s
41 DeepSeek V4 Flash 0731 32.72% $0.37 9.12m 27m43s
42 Beam 1.0 22.83% - 3.09m 7m13s
43 Muse Spark 1.2 21.78% $2.64 10.55m 25m35s
44 Inkling 14.94% $1.71 6.86m 30m12s
45 Inkling Small 9.33% $0.34 3.55m 11m18s
46 Mercury 2.5 2.44% $0.16 1.91m 4m03s

Why IOI?

Recently, top LLM labs like OpenAI and Google reported that their models achieved gold medals on the International Mathematical Olympiad (IMO). However, advanced models are starting to saturate IMO, meaning it may no longer effectively differentiate between the capabilities of top-performing models. Reports also suggest the evaluation process faced coordination challenges, with AI companies seeking expedited validation mid-competition that may not reflect standard IMO assessment procedures.

The International Olympiad in Informatics (IOI) offers several advantages as an LLM benchmark. The competition features standardized and automated grading, ensuring objective evaluation without subjective scoring. Additionally, the IOI has real-world relevance as it tests C++ programming skills that are directly applicable to software development.


Benchmark Design

We designed our benchmark to imitate competition conditions as closely as possible.

Agent Harness

Each model runs inside OpenCode, a general-purpose coding agent, in an isolated sandbox with a c++ (v20) toolchain. Its workspace holds exactly what the contest hands a contestant: the problem statement, the task header and solution stub, the sample grader or manager, the compile and run scripts, and the sample input and output from the statement. The official test data, the subtask test lists, and the grading script are withheld until the attempt is over. The agent reads, writes, compiles and tests files freely in its workspace, and writes its answer to /workspace/solution.cpp. Like a contestant, it has no access to the public internet: the sandbox reaches only our model gateway, so published editorials and reference solutions are out of reach.

Unlike a contestant, the agent has no submission tool and therefore receives no score feedback during the attempt: it can run the samples and whatever tests it writes for itself, and must decide on its own when its solution is good enough.

Some models spend their entire per-response output budget reasoning about a problem before writing any code. When that happens, the harness continues the cut-off turn on the next step rather than failing the problem: it replays the provider’s own reasoning state where the API returns one, and otherwise restarts the turn with an instruction to work in smaller steps. The same rule applies to every model on this board, and the cost column includes the reasoning spent this way.

Results from our earlier harness, which gave the agent an interactive grading tool, are preserved on IOI v1. The two harnesses are not comparable, so their results are reported on separate boards.

Scoring

Grading matches the olympiad: after the attempt ends, the solution left in the workspace is copied into a fresh grading directory alongside pristine copies of the official test data and grader, compiled, run against every official test, and scored per subtask out of 100 points. Every grading file is verified unchanged after the run. A solution that fails to compile, or that the agent never wrote, scores zero. A model’s score for a year is its mean score across that year’s six problems, and its overall score is the mean of its three yearly scores.


Results

Gemini 4 Argon (100.00%), GPT-6 Astra (100.00%), and GPT-6.1 Sol (96.89%) lead the board. Claude Sonnet 5 is expensive: on thirteen of its eighteen problems it spent its entire 128k-token output budget reasoning before writing any code, and the harness had to continue the cut-off turn.

We evaluate three years to check for data contamination. The 2026 problems were released only after the models in this cohort were trained, and they are the hardest set for many of them. Gemini 4 Argon scores 100.00% on 2026, so the newest set is not out of reach.