OJBench

Gemini 3 Pro 70% leads since November 2025

Keye-VL-2.0-30B-A3B 71.5% leads since June 2026

Fading

Official leaderboard Include claimed scores

Models write C++ or Python solutions to 232 competition problems from NOI and ICPC. Judge test cases score each solution by pass@1.

Made by: Beijing University of Posts and Telecommunications, Tsinghua University, and others
Released: June 2025
Kind: Competitive programming
Level: Competition
Scored: Checked automatically
Answers: Code
Tasks: All public

Strongest model tested: Claude Opus 4.6 #37 on the capability index
Last in a launch post: May 2026, dropped before it saturated
Last model added: June 2026
Scores refreshed: November 12th, 2026

In the authors’ words

Recent advancements in large language models (LLMs) have demonstrated significant progress in math and code reasoning capabilities. However, existing code benchmark are limited in their ability to evaluate the full spectrum of these capabilities, particularly at the competitive level. To bridge this gap, we introduce OJBench, a novel and challenging benchmark designed to assess the competitive-level code reasoning abilities of LLMs. OJBench comprises 232 programming competition problems from NOI and ICPC, providing a more rigorous test of models' reasoning skills. We conducted a comprehensive evaluation using OJBench on 37 models, including both closed-source and open-source models, reasoning-oriented and non-reasoning-oriented models. Our results indicate that even state-of-the-art reasoning-oriented models, such as o4-mini and Gemini-2.5-pro-exp, struggle with highly challenging competition-level problems. This highlights the significant challenges that models face in competitive-level code reasoning.

Abstract of OJBench: A Competition Level Code Benchmark For Large Language Models (arXiv:2506.16395, CC0)

Models scored on its own leaderboard

Launch posts list it, from 4 labs

System cards report it, from 4 labs

Gemini 3 Pro leads with 70% among models released by February 2026

Best score per model from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best.

Keye-VL-2.0-30B-A3B leads with 71.5%, claims included

Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model.

Bars Table

The newest model with a score here came out in February 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

Source: Official OJBench leaderboard

1 Keye-VL-2.0-30B-A3B 71.5%
2 Gemini 3 Pro 70%
3 Claude Opus 4.6 thinking 67.2%
4 Kimi K2.6 60.6%
5 Kimi K2.5 57.4%
6 GLM-5 54.9%
7 Seed2.0 Pro 54.9%
8 DeepSeek-V3.2 51.7%
9 GLM-4.7 50.8%
10 Kimi K2 Thinking 48.7%
11 LongCat-Flash-Thinking-2601 42.2%
12 MiniMax-M2.5 41.4%
13 LongCat-Flash-Thinking 40.7%
14 Qwen3.5-27B 40.1%
15 Qwen3.5-122B-A10B 39.5%
16 Qwen3.5-35B-A3B 36%
17 Qwen3-235B-A22B-Thinking Jul 2025 32.5%
18 Qwen3-Next-80B-A3B 29.7%
19 Qwen3.5-9B 29.2%
20 Qwen3-VL-235B-A22B 27.5%
21 Kimi K2 Jul 2025 27.1%
22 Qwen3.5-4B 24.1%
23 Qwen3-Coder-Next 23%

Sources: Labs’ model cards and official OJBench leaderboard

#	Model	Score	Setting	Date	Source
1	Gemini 3 Pro	70%		Nov 2025	Official leaderboard
2	Claude Opus 4.6	67.2%	thinking	Feb 2026	Official leaderboard
3	Kimi K2.5	57%	thinking	Jan 2026	Official leaderboard
4	GLM-5	54.9%		Feb 2026	Official leaderboard
5	Seed2.0 Pro	54.9%		Feb 2026	Official leaderboard
6	DeepSeek-V3.2	51.7%		Dec 2025	Official leaderboard
7	GLM-4.7	50.8%		Dec 2025	Official leaderboard
8	MiniMax-M2.5	41.4%		Feb 2026	Official leaderboard

#	Model	Score	Setting	Date	Source
1	Keye-VL-2.0-30B-A3B	71.5%		Jun 2026	Lab model card
2	Gemini 3 Pro	70%		Nov 2025	Official leaderboard
3	Claude Opus 4.6	67.2%	thinking	Feb 2026	Official leaderboard
4	Kimi K2.6	60.6%		Apr 2026	Lab model card
5	Kimi K2.5	57.4%		Jan 2026	Lab model card
6	GLM-5	54.9%		Feb 2026	Official leaderboard
7	Seed2.0 Pro	54.9%		Feb 2026	Official leaderboard
8	DeepSeek-V3.2	51.7%		Dec 2025	Official leaderboard
9	GLM-4.7	50.8%		Dec 2025	Official leaderboard
10	Kimi K2 Thinking	48.7%		Nov 2025	Lab model card
11	LongCat-Flash-Thinking-2601	42.2%		Jan 2026	Lab model card
12	MiniMax-M2.5	41.4%		Feb 2026	Official leaderboard
13	LongCat-Flash-Thinking	40.7%		Sep 2025	Lab model card
14	Qwen3.5-27B	40.1%		Feb 2026	Lab model card
15	Qwen3.5-122B-A10B	39.5%		Feb 2026	Lab model card
16	Qwen3.5-35B-A3B	36%		Feb 2026	Lab model card
17	Qwen3-235B-A22B-Thinking Jul 2025	32.5%		Jul 2025	Lab model card
18	Qwen3-Next-80B-A3B	29.7%		Sep 2025	Lab model card
19	Qwen3.5-9B	29.2%		Mar 2026	Lab model card
20	Qwen3-VL-235B-A22B	27.5%		Nov 2025	Lab model card
21	Kimi K2 Jul 2025	27.1%		Jul 2025	Lab model card
22	Qwen3.5-4B	24.1%		Mar 2026	Lab model card
23	Qwen3-Coder-Next	23%		Feb 2026	Lab model card

Labs report scores for 16 models in their cards. Switch the scores above to rank them together.

Number one over time 6 models have held number one with claims included since its release in June 2025

Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.

Google DeepMind Moonshot AI Qwen Meituan Kuaishou

Own leaderboard Not tested since Feb 2026

Gemini 3 Pro 70%

Lab claims Not tested since Jun 2026

Keye-VL-2.0-30B-A3B 71.5%

Sources: Official OJBench leaderboard and labs’ model cards

Scores over time

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

On the benchmark’s own leaderboard Reported by the lab in its model card Best so far

Sources: Official OJBench leaderboard and labs’ model cards

Labs headline it less often now, last in May 2026

Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.

Moonshot AI

Qwen

Meituan

Kuaishou

Date	Launch post	Score	As the post writes it
May 25, 2026	Keye-VL-2.0-30B-A3B	39.2	OJBench
Apr 20, 2026	Kimi K2.6	60.6	OJBench (python)
Mar 2, 2026	Qwen3.5-4B	24.1	OJBench
Mar 2, 2026	Qwen3.5-9B	29.2	OJBench
Feb 24, 2026	Qwen3.5-35B-A3B	36	OJBench
Feb 24, 2026	Qwen3.5-27B	40.1	OJBench
Feb 24, 2026	Qwen3.5-122B-A10B	39.5	OJBench
Jan 27, 2026	Kimi K2.5	57.4	OJBench (cpp)
Jan 14, 2026	LongCat-Flash-Thinking-2601	42.2	OJBench (Pass@1)
Nov 6, 2025	Kimi K2 Thinking	48.7	OJ-Bench (cpp), no tools
Sep 21, 2025	LongCat-Flash-Thinking	40.7	OJBench(Mean@1)
Sep 11, 2025	Qwen3-Next-80B-A3B	29.7	OJBench
Jul 25, 2025	Qwen3-235B-A22B-Thinking	32.5	OJBench
Jul 11, 2025	Kimi K2	27.1	OJBench, Pass@1

7 system cards report it, from 4 labs

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

Date	Document	Score	As the document writes it	Section
Jun 9, 2026	Keye-VL-2.0-30B-A3B technical report	71.5	OJBench (Table 3 and Section 6.2.1 text). Figure 5 prints 39.2 for the same row	Capabilities
Feb 28, 2026	Qwen3-Coder-Next technical report	23	OJBench (Table 6)	Capabilities
Jan 30, 2026	Kimi K2.5 technical report	57.4	OJBench (cpp) (Table 4)	Capabilities
Jan 23, 2026	LongCat-Flash-Thinking-2601 technical report	42.2	OJBench (Pass@1) (Table 2)	Capabilities
Nov 26, 2025	Qwen3-VL-235B-A22B technical report	27.5	OJBench (Table 6)	Capabilities
Sep 23, 2025	LongCat-Flash-Thinking technical report	40.7	OJBench(Mean@1) (Table 2)	Capabilities
Jul 21, 2025	Kimi K2 technical report	27.1	OJBench (Pass@1) (Table 3)	Capabilities

One lab claim next to the leaderboard

Each score a lab reported for its own model, beside the same model's best score on the leaderboard. Gaps mostly come from different harnesses, effort levels and task subsets. Every benchmark.

Model	Lab says	Board says	Gap
Kimi K2.5	57.4% OJBench (cpp)	57% Leaderboard, thinking	+0.4 pts