The 2026 gold medal threshold was 29 points (down from 35 in 2025, reflecting harder problems), and dots-note-3.0 exceeded it by 13 points . For context, only 7 out of 666 human contestants — just over 1% — achieved a perfect score this year .

Multiple reports also note that Huawei separately claimed a 100% score on the same problems, making 2026 the first year two AI systems achieved perfect marks .

How this compares to prior AI performances at the IMO

The progression has been rapid:

• 2024: Google DeepMind AlphaProof scored 28 out of 42 (silver) and was the first AI to prove answers with formally verified correctness via the Lean proof assistant . It solved 4 of 6 problems, one point short of gold.
• 2025: Both Google DeepMind's Gemini Deep Think and an OpenAI experimental reasoning model achieved gold-level scores of 35 out of 42, solving 5 of 6 problems . This was the first time AI reached gold territory.
• 2026: RedNote's dots-note-3.0 is the first to achieve a perfect 42/42 — the first time any AI has solved every problem . It did so using a comparatively lightweight model, and Huawei also reported a perfect score in the same cycle .

A quick summary:

The credibility and verification context

A critical caveat matters: dots-note-3.0 did not sit the official IMO competition alongside the 666 human contestants . The IMO's General Regulations restrict participation to invited countries and human contestants under age 20 . Instead, tech firms are provided with the same problems after human contestants have completed the exam, and their solutions are judged by IMO graders .

In 2025, the IMO committee did not write formal rules for AI participation, which meant each company could define its own testing conditions — leading mathematicians to note this enabled companies to "both set and mark their own homework" . The same informal process applies in 2026. As one analysis put it, dots-note-3.0 "did not enter the competition in the traditional sense" — it solved the problems post-hoc under company-defined testing protocols .

That said, RedNote stated that its model's solutions were submitted for official IMO grading and awarded full points by the same judges who score human contestants . Huawei's model was similarly verified. This provides a meaningful layer of credibility, since the graders are independent and the solutions were judged against the same rubric.

What this means

The achievement is genuine in the sense that the model produced correct, IMO-graded solutions to all six problems. But it is not an official IMO medal — the IMO does not formally award medals to AI systems. The comparison to prior AI results is fair because those systems also solved the same problems post-hoc under similar conditions, but all IMO "AI gold medals" should be understood as unofficial, retrospectively judged benchmarks rather than direct wins in a live competition.

The dots-note-3.0 result is nevertheless a significant milestone as the first time any AI has reached a perfect score on a full IMO problem set — a sign that AI mathematical reasoning continues to advance at a remarkable pace, even if the precise meaning of "winning" the IMO remains contested .

The 2026 gold medal threshold was 29 points (down from 35 in 2025, reflecting harder problems), and dots-note-3.0 exceeded it by 13 points . For context, only 7 out of 666 human contestants — just over 1% — achieved a perfect score this year .

Multiple reports also note that Huawei separately claimed a 100% score on the same problems, making 2026 the first year two AI systems achieved perfect marks .

How this compares to prior AI performances at the IMO

The progression has been rapid:

• 2024: Google DeepMind AlphaProof scored 28 out of 42 (silver) and was the first AI to prove answers with formally verified correctness via the Lean proof assistant . It solved 4 of 6 problems, one point short of gold.
• 2025: Both Google DeepMind's Gemini Deep Think and an OpenAI experimental reasoning model achieved gold-level scores of 35 out of 42, solving 5 of 6 problems . This was the first time AI reached gold territory.
• 2026: RedNote's dots-note-3.0 is the first to achieve a perfect 42/42 — the first time any AI has solved every problem . It did so using a comparatively lightweight model, and Huawei also reported a perfect score in the same cycle .

A quick summary:

The credibility and verification context

A critical caveat matters: dots-note-3.0 did not sit the official IMO competition alongside the 666 human contestants . The IMO's General Regulations restrict participation to invited countries and human contestants under age 20 . Instead, tech firms are provided with the same problems after human contestants have completed the exam, and their solutions are judged by IMO graders .

In 2025, the IMO committee did not write formal rules for AI participation, which meant each company could define its own testing conditions — leading mathematicians to note this enabled companies to "both set and mark their own homework" . The same informal process applies in 2026. As one analysis put it, dots-note-3.0 "did not enter the competition in the traditional sense" — it solved the problems post-hoc under company-defined testing protocols .

That said, RedNote stated that its model's solutions were submitted for official IMO grading and awarded full points by the same judges who score human contestants . Huawei's model was similarly verified. This provides a meaningful layer of credibility, since the graders are independent and the solutions were judged against the same rubric.

What this means

The achievement is genuine in the sense that the model produced correct, IMO-graded solutions to all six problems. But it is not an official IMO medal — the IMO does not formally award medals to AI systems. The comparison to prior AI results is fair because those systems also solved the same problems post-hoc under similar conditions, but all IMO "AI gold medals" should be understood as unofficial, retrospectively judged benchmarks rather than direct wins in a live competition.

The dots-note-3.0 result is nevertheless a significant milestone as the first time any AI has reached a perfect score on a full IMO problem set — a sign that AI mathematical reasoning continues to advance at a remarkable pace, even if the precise meaning of "winning" the IMO remains contested .