# Mellum2-12B-A2.5B-Thinking vs Qwen3.8-27B

> Side-by-side benchmark comparison for Mellum2-12B-A2.5B-Thinking and Qwen3.8-27B across agentic, coding, multimodal, knowledge, reasoning, multilingual, and math tasks.

- Canonical page: https://benchlm.ai/compare/mellum2-12b-a2-5b-thinking-vs-qwen3-8-27b
- Last updated: September 28, 2026

- Shared sourced benchmarks: 3
- HTML indexing: indexable
- Ranking lane: BenchAlign v5.7

## Quick Verdict

Evidence is too thin for a benchmark verdict: Mellum2-12B-A2.5B-Thinking and Qwen3.8-27B share only 3 sourced benchmark rows.

## Summary

- Only 3 shared sourced benchmark rows are available. That is not enough for a broad capability verdict, so this mirror does not name an overall winner.
- Qwen3.8-27B has the larger published context window at 262K.

## Model Snapshot

| Property | Mellum2-12B-A2.5B-Thinking | Qwen3.8-27B |
|----------|----------|----------|
| Creator | JetBrains | Alibaba |
| Type | Open Weight | Open Weight |
| Reasoning | Reasoning | Reasoning |
| Context | 128K | 262K |
| Independent public score (not head-to-head) | null | 55.26 |
| Benchmarks Covered | 6 | 33 |
| Pricing (input/output) | N/A | $0.00 / $0.00 |

## Category Breakdown

### Agentic

- Winner: Insufficient shared evidence
- Mellum2-12B-A2.5B-Thinking public-lane score: Coming soon
- Qwen3.8-27B public-lane score: 61.2 (Supported · #18/117)

| Benchmark | Mellum2-12B-A2.5B-Thinking | Qwen3.8-27B | Winner |
|-----------|-----------|-----------|--------|
| BFCL v4 | 45.6% | Coming soon | Coming soon |
| Terminal-Bench 2.1 | Coming soon | 73.0% | Coming soon |
| CoWorkBench | Coming soon | 70.7% | Coming soon |
| JobBench | Coming soon | 33.4% | Coming soon |
| Agents' Last Exam | Coming soon | 42.9% | Coming soon |
| OSWorld-Verified | Coming soon | 84.3% | Coming soon |
| WebArena-Verified | Coming soon | 64.8% | Coming soon |
| AndroidWorld | Coming soon | 81.9% | Coming soon |
| Terminal-Bench 2.1 (Vals) | Coming soon | 58.4% | Coming soon |

### Coding

- Winner: Insufficient shared evidence
- Mellum2-12B-A2.5B-Thinking public-lane score: Coming soon
- Qwen3.8-27B public-lane score: 48.7 (Supported · #45/142)

| Benchmark | Mellum2-12B-A2.5B-Thinking | Qwen3.8-27B | Winner |
|-----------|-----------|-----------|--------|
| LiveCodeBench v6 | 69.9% | 90.3% | Qwen3.8-27B |
| Terminal-Bench 2.1 | Coming soon | 73.0% | Coming soon |
| SWE-bench Pro | Coming soon | 61.7% | Coming soon |
| NL2Repo | Coming soon | 42.3% | Coming soon |
| DeepSWE | Coming soon | 42.2% | Coming soon |
| VulcanBench v3 | Coming soon | 82.6% | Coming soon |
| LiveCodeBench (Vals) | Coming soon | 84.0% | Coming soon |
| SWE-bench (Vals) | Coming soon | 86.0% | Coming soon |

### Reasoning

- Winner: Insufficient shared evidence
- Mellum2-12B-A2.5B-Thinking public-lane score: Coming soon
- Qwen3.8-27B public-lane score: 78.7 (#8/27)

### Multimodal & Grounded

- Winner: Insufficient shared evidence
- Mellum2-12B-A2.5B-Thinking public-lane score: Coming soon
- Qwen3.8-27B public-lane score: 80.9 (#11/50)

| Benchmark | Mellum2-12B-A2.5B-Thinking | Qwen3.8-27B | Winner |
|-----------|-----------|-----------|--------|
| MathVision | Coming soon | 90.0% | Coming soon |
| MathVision w/ Python | Coming soon | 94.6% | Coming soon |
| BabyVision | Coming soon | 65.7% | Coming soon |
| BabyVision w/ Python | Coming soon | 85.6% | Coming soon |
| Vision2Web | Coming soon | 62.9% | Coming soon |
| CharXiv w/o tools | Coming soon | 83.7% | Coming soon |
| CharXiv | Coming soon | 90.2% | Coming soon |
| OmniDocBench 1.5 | Coming soon | 91.1% | Coming soon |
| RealWorldQA | Coming soon | 85.9% | Coming soon |
| ERQA | Coming soon | 65.5% | Coming soon |

### Knowledge

- Winner: Insufficient shared evidence
- Mellum2-12B-A2.5B-Thinking public-lane score: Coming soon
- Qwen3.8-27B public-lane score: 49.2 (Supported · #63/168)

| Benchmark | Mellum2-12B-A2.5B-Thinking | Qwen3.8-27B | Winner |
|-----------|-----------|-----------|--------|
| MMLU-Redux | 86.2% | Coming soon | Coming soon |
| GPQA | 57.6% | 89.2% | Qwen3.8-27B |
| GPQA-D | 57.6% | 89.2% | Qwen3.8-27B |
| HLE | Coming soon | 30.8% | Coming soon |
| HLE w/o tools | Coming soon | 30.8% | Coming soon |
| GPQA Diamond (Vals) | Coming soon | 88.9% | Coming soon |
| MMLU-Pro (Vals) | Coming soon | 84.3% | Coming soon |

### Instruction Following

- Winner: Insufficient shared evidence
- Mellum2-12B-A2.5B-Thinking public-lane score: Coming soon
- Qwen3.8-27B public-lane score: 83.2 (#45/124)

| Benchmark | Mellum2-12B-A2.5B-Thinking | Qwen3.8-27B | Winner |
|-----------|-----------|-----------|--------|
| IFEval | 76.5% | Coming soon | Coming soon |
| IFBench | Coming soon | 79.5% | Coming soon |

## FAQ

### Is there enough evidence to pick between Mellum2-12B-A2.5B-Thinking and Qwen3.8-27B?

Not yet. The models share only 3 sourced benchmark rows, so this mirror reports the available measurements and metadata without naming a broad winner.

### What can I compare safely on this page?

Published pricing, context windows, model type, and individual shared benchmark rows are directly comparable. Independent overall scores and non-overlapping benchmark coverage are labeled separately and should not be read as a controlled head-to-head result.

## Related Comparisons

- [Mellum2-12B-A2.5B-Thinking vs Mellum2-12B-A2.5B-Instruct](/compare/mellum2-12b-a2-5b-instruct-vs-mellum2-12b-a2-5b-thinking)
- [Qwen3.8-27B vs Mellum2-12B-A2.5B-Instruct](/compare/mellum2-12b-a2-5b-instruct-vs-qwen3-8-27b)
- [Mellum2-12B-A2.5B-Thinking vs GPT-6 Astra](/compare/gpt-6-astra-vs-mellum2-12b-a2-5b-thinking)
- [Qwen3.8-27B vs GPT-6 Astra](/compare/gpt-6-astra-vs-qwen3-8-27b)

## Explore More

- [Mellum2-12B-A2.5B-Thinking profile](/models/mellum2-12b-a2-5b-thinking)
- [Qwen3.8-27B profile](/models/qwen3-8-27b)
- [Compare Pricing](/llm-pricing)
- [Alternative Finder](/tools/alternative-finder)
- [LLM Selector](/tools/llm-selector)
- [Overall Rankings](/best/overall)