Qwen3.8-Flash-Next vs Flint-40B-A2.5B vs Qwen3.8-27B
New ComparisonScores are fetched live from the current database on every visit — nothing is cached.
Winner
Flint-40B-A2.5B scores higher overall across category strength profiles (63 vs 63 for Qwen3.8-27B); only 1 benchmarks are directly comparable.
Only 1 of 50 benchmarks are scored by every selected model — the verdict leans on category strength profiles and is indicative.
Benchmark Scores Comparison
Qwen3.8-Flash-Next
125.0B params
Score: 70.0
Flint-40B-A2.5B
39.8B params
Qwen3.8-27B
27.0B params
Score: 100.0
Coverage matrix
3 models × 50 benchmarks — hover a benchmark name for its category
| Benchmark | Qwen3.8-27B Qwen Team (Alibaba Cloud) | Qwen3.8-Flash-Next Qwen Team (Alibaba) | Flint-40B-A2.5B Recursive |
|---|---|---|---|
| GPQA Diamond | 89.2 | 91.7 | — |
| IFBench (prompt loose) | 79.5 | — | — |
| LiveCodeBench v6 | 90.3 | 91.9 | — |
| SWE-bench Pro | 61.7 | 62.5 | 89.7 |
| BFCL-V4 | 72.4 | — | — |
| RealWorldQA | 85.9 | 88.5 | — |
| Agents' Last Exam (Score) | 42.9 | 51.2 | — |
| AA-LCR | 80.3 | — | — |
| MathVision | 90 | 90.6 | — |
| AIME 26 | 98.8 | — | — |
| AndroidWorld | 81.9 | 84.5 | — |
| Agents' Last Exam | 20.4 | 24.3 | — |
| Toolathlon Verified | 67.1 | 73.5 | — |
| RecreationBench | 47.1 | 49.9 | — |
| AutomationBench (Public) | 58.7 | — | — |
| GDPVal | 63.5 | — | — |
| NL2Repo | 42.3 | — | — |
| BabyVision | 65.7 | — | — |
| Claw-Eval Avg | 56.9 | 60.4 | — |
| Multi-Challenge | 63.6 | — | — |
| SWE-MM | 38.6 | — | — |
| MMLU-Pro | 85.7 | — | — |
| OSWorld-Verified | 84.3 | — | — |
| HLE (with tools) | 30.8 | 35.9 | — |
| LVBench | 72.4 | 76.6 | — |
| CharXiv (RQ) | 83.7 | 84.6 | — |
| GDP.pdf | 11.2 | — | — |
| NL2Repo-Bench | 42.3 | 48.1 | — |
| WebArena-Verified | 64.8 | — | — |
| CoWorkBench | 70.7 | 73.9 | — |
| AA-Omniscience Index (public set) | -0.5 | — | — |
| SWE-bench Multilingual | 73.8 | 81 | — |
| IFBench | 79.5 | 81.3 | — |
| SWE-bench Verified | 74.6 | — | 74.2 |
| Humanity's Last Exam | 30.8 | — | — |
| ERQA | 65.5 | 72.3 | — |
| QwenSWEBench | 79 | — | — |
| Terminal-Bench 2.1 (Terminus-2) | 73 | — | 54.9 |
| OmniDocBench 1.5 | 91.1 | — | — |
| HMMT Feb 26 | 95.8 | — | — |
| Terminal-Bench Lite | 67 | — | — |
| JobBench | 33.4 | 55.7 | — |
| OSWorld 2.0 (Partial) | 48 | 52.3 | — |
| OSWorld 2.0 (Binary) | 19.4 | 19.4 | — |
| Vision2Web | 62.9 | 64 | — |
| Claw-Eval Pass^3 | 57.4 | 64.4 | — |
| DeepSWE 1.1 | 42.2 | 58.7 | — |
| Chatbot Korean (internal) | 90.1 | — | — |
| SciCode | 44.6 | — | — |
| BrowseComp Plus | 58 | — | — |
| Covered | 50/50 | 25/50 | 3/50 |
Scorecard
Overall 0–100 score — category strengths (50%), shared-benchmark wins (30%), coverage (20%)
| Model | Overall score | Shared benchmark wins | Coverage | Strongest at | Weakest at |
|---|---|---|---|---|---|
| Flint-40B-A2.5B Winner | 62.9 | 1 of 1 | 3 of 50 (6.0%) | Agentic Coding | Agentic Coding |
| Qwen3.8-27B | 62.6 | 0 of 1 | 50 of 50 (100.0%) | Instruction Following | General Agents |
| Qwen3.8-Flash-Next | 57.7 | 0 of 1 | 25 of 50 (50.0%) | Vision & Language | Agentic Coding |
Category strengths
Where each model stands, per benchmark category — score relative to the best model, adjusted for benchmark coverage. This also ranks models when they share few benchmarks.
| Category | Flint-40B-A2.5B | Qwen3.8-27B | Qwen3.8-Flash-Next |
|---|---|---|---|
| General Agents 15 benchmarks | — | 77.3 | 100 |
| Agentic Coding 12 benchmarks | 63.4 | 85.7 | 84.3 |
| STEM & Reasoning 6 benchmarks | — | 89.9 | 100 |
| Vision & Language 4 benchmarks | — | 95.9 | 100 |
| Instruction Following 3 benchmarks | — | 96.9 | 100 |
| Document Understanding 2 benchmarks | — | 95.2 | 100 |
| Spatial Intelligence 1 benchmark | — | 79.9 | 100 |
| Video Understanding 1 benchmark | — | 88.1 | 100 |
Fully comparable benchmarks
Scored by all 3 selected models (1)
| Benchmark | Category | Qwen3.8-Flash-Next | Flint-40B-A2.5B | Qwen3.8-27B |
|---|---|---|---|---|
| SWE-bench Pro | coding_agent | 62.5 | 89.7 | 61.7 |
Partial coverage benchmarks
Scored by at least one but not all selected models (49) — missing scores show as —
| Benchmark | Category | Qwen3.8-Flash-Next | Flint-40B-A2.5B | Qwen3.8-27B |
|---|---|---|---|---|
| GPQA Diamond 2/3 models | stem_reasoning | 91.7 | — | 89.2 |
| IFBench (prompt loose) 1/3 models | instruction_following | — | — | 79.5 |
| LiveCodeBench v6 2/3 models | stem_reasoning | 91.9 | — | 90.3 |
| BFCL-V4 1/3 models | general_agent | — | — | 72.4 |
| RealWorldQA 2/3 models | vision_language | 88.5 | — | 85.9 |
| Agents' Last Exam (Score) 2/3 models | general_agent | 51.2 | — | 42.9 |
| AA-LCR 1/3 models | long_context | — | — | 80.3 |
| MathVision 2/3 models | vision_language | 90.6 | — | 90 |
| AIME 26 1/3 models | stem_reasoning | — | — | 98.8 |
| AndroidWorld 2/3 models | general_agent | 84.5 | — | 81.9 |
| Agents' Last Exam 2/3 models | general_agent | 24.3 | — | 20.4 |
| Toolathlon Verified 2/3 models | general_agent | 73.5 | — | 67.1 |
| RecreationBench 2/3 models | general_agent | 49.9 | — | 47.1 |
| AutomationBench (Public) 1/3 models | general_agent | — | — | 58.7 |
| GDPVal 1/3 models | general_agent | — | — | 63.5 |
| NL2Repo 1/3 models | coding_agent | — | — | 42.3 |
| BabyVision 1/3 models | vision_language | — | — | 65.7 |
| Claw-Eval Avg 2/3 models | coding_agent | 60.4 | — | 56.9 |
| Multi-Challenge 1/3 models | instruction_following | — | — | 63.6 |
| SWE-MM 1/3 models | coding_agent | — | — | 38.6 |
| MMLU-Pro 1/3 models | knowledge | — | — | 85.7 |
| OSWorld-Verified 1/3 models | general_agent | — | — | 84.3 |
| HLE (with tools) 2/3 models | stem_reasoning | 35.9 | — | 30.8 |
| LVBench 2/3 models | video_understanding | 76.6 | — | 72.4 |
| CharXiv (RQ) 2/3 models | document_understanding | 84.6 | — | 83.7 |
| GDP.pdf 1/3 models | long_context | — | — | 11.2 |
| NL2Repo-Bench 2/3 models | coding_agent | 48.1 | — | 42.3 |
| WebArena-Verified 1/3 models | general_agent | — | — | 64.8 |
| CoWorkBench 2/3 models | general_agent | 73.9 | — | 70.7 |
| AA-Omniscience Index (public set) 1/3 models | knowledge | — | — | -0.5 |
| SWE-bench Multilingual 2/3 models | coding_agent | 81 | — | 73.8 |
| IFBench 2/3 models | instruction_following | 81.3 | — | 79.5 |
| SWE-bench Verified 2/3 models | coding_agent | — | 74.2 | 74.6 |
| Humanity's Last Exam 1/3 models | stem_reasoning | — | — | 30.8 |
| ERQA 2/3 models | spatial_intelligence | 72.3 | — | 65.5 |
| QwenSWEBench 1/3 models | coding_agent | — | — | 79 |
| Terminal-Bench 2.1 (Terminus-2) 2/3 models | coding_agent | — | 54.9 | 73 |
| OmniDocBench 1.5 1/3 models | document_understanding | — | — | 91.1 |
| HMMT Feb 26 1/3 models | stem_reasoning | — | — | 95.8 |
| Terminal-Bench Lite 1/3 models | coding_agent | — | — | 67 |
| JobBench 2/3 models | general_agent | 55.7 | — | 33.4 |
| OSWorld 2.0 (Partial) 2/3 models | general_agent | 52.3 | — | 48 |
| OSWorld 2.0 (Binary) 2/3 models | general_agent | 19.4 | — | 19.4 |
| Vision2Web 2/3 models | vision_language | 64 | — | 62.9 |
| Claw-Eval Pass^3 2/3 models | coding_agent | 64.4 | — | 57.4 |
| DeepSWE 1.1 2/3 models | coding_agent | 58.7 | — | 42.2 |
| Chatbot Korean (internal) 1/3 models | multilingual | — | — | 90.1 |
| SciCode 1/3 models | — | — | — | 44.6 |
| BrowseComp Plus 1/3 models | general_agent | — | — | 58 |
Comparison FAQ
Which model scores highest in this comparison?
Flint-40B-A2.5B scores higher overall across category strength profiles (63 vs 63 for Qwen3.8-27B); only 1 benchmarks are directly comparable.
Which model is strongest at General Agents?
Qwen3.8-Flash-Next leads the General Agents category (strength score 100/100 across 9 comparable benchmarks).
How reliable is this comparison verdict?
Only 1 of 50 benchmarks are scored by every selected model — the verdict leans on category strength profiles and is indicative.