Mellum2-12B-A2.5B-Instruct-SFT: benchmark scores, price and ranking

Mellum2-12B-A2.5B-Instruct-SFT from JetBrains ranks #509 of 743 LLMs on the LLMs Tiger aggregate benchmark score (46.5), based on 1 benchmark results captured up to 2026-10-13.

Organisation: JetBrains
Type: Open-weight / local model
Open weights: Yes
Aggregate rank: #509 of 743 (score 46.5)

Across the 1 benchmark where at least ten models are scored, Mellum2-12B-A2.5B-Instruct-SFT is in the top 10% on 0 and in the top half on 0.

Mellum2-12B-A2.5B-Instruct-SFT benchmark results

Mellum2-12B-A2.5B-Instruct-SFT benchmark scores, ranked against all models measured on each benchmark
Benchmark	Score	Rank	Source	Captured
GPQA Diamond	38.9%	#225 of 275	HuggingFace GPQA Diamond leaderboard	2026-10-13

About Mellum2-12B-A2.5B-Instruct-SFT

Mellum2-12B-A2.5B-Instruct-SFT is an open-weight model that can run on your own hardware from JetBrains. No API price is listed for it yet. We hold 1 benchmark result for it across 1 area, and none is in the top tenth of its board.

Where Mellum2-12B-A2.5B-Instruct-SFT stands

On our aggregate score, Mellum2-12B-A2.5B-Instruct-SFT is #509 of 743 ranked models, in the lower half. Its score is 46.5; the scale is centred on 50, so it sits 3.5 points below the typical model. That beats 31% of the ranked field. Among the 6 ranked JetBrains models it is #4. The next one up is Mellum2-12B-A2.5B-Thinking-SFT at 46.6. The next one down is Mellum2-12B-A2.5B-Base at 44.0.

Mellum2-12B-A2.5B-Instruct-SFT, benchmark by benchmark

GPQA Diamond is a weak spot. Mellum2-12B-A2.5B-Instruct-SFT scores 38.9%, #225 of 275, behind 81% of the models measured. The leader, GPT-6 Astra (max), reaches 95.8%. It covers expert-level science questions. The board median is 71.7%, so Mellum2-12B-A2.5B-Instruct-SFT is 32.8 points below it. Source: HuggingFace GPQA Diamond leaderboard, captured 2026-10-13.

Reading these numbers

The 1 score on this page comes from 1 source: HuggingFace GPQA Diamond leaderboard (1). No official leaderboard supplied any of them, so treat them as third-party measurements. With 1 results the aggregate rank can move a lot when one new score arrives.

Models ranked nearby

#506 qwen3.6-plus (46.6)
#507 Phi-3.5-mini-instruct (46.5)
#508 grok-3 (46.5)
#510 Qwen2.5-Coder-3B-Instruct (46.5)
#511 amazon.nova-pro-v1:0 (46.5)
#512 aya-expanse-32b (46.5)