Across 1 comparable benchmark, MiniMax M3.1 Flash has better reported results on 1 and worse on 0.
Limited evidence1 of 1 shared result countedHow we compare
Price basis: Listed rates vs standard rates. The price section below shows the applicable periods.
| Fact | Qwen3-VL-4B-Instruct | MiniMax M3.1 Flash |
|---|---|---|
| Context window | 256K tokens | 1M tokens |
| Input / 1M | — | $0.150 |
| Output / 1M | — | $0.600 |
| Blended (4:1) | — | $0.143 |
| Status | Available | Available |
| Released | October 15, 2025 | October 19, 2026 |
Blended is 4:1 input to output with 90% cache hits at listed rates — set your own mix.
Mean of 1 shared benchmark grouped by capability on a 0–100 scale. How averages are built
MiniMax M3.1 Flash ahead on 1 of 1 counted
No shared comparable results for Coding, Abstract reasoning, Mathematics, Science & health, Computer & tool use, Research & search, Long context, Professional work & productivity, Cybersecurity.
Published scores, with matching modes where available. Gaps show Qwen3-VL-4B-Instruct relative to MiniMax M3.1 Flash. Reading the results
| Benchmark | Qwen3-VL-4B-Instruct | MiniMax M3.1 Flash | Gap |
|---|---|---|---|
| MMMU-ProVisual Understanding · Accuracy | 43.9Normal · no tools | 79.1≠Thinking · no tools | −35.2 pts#129 vs #37 of 136 |
Visual Understanding · Accuracy
A dash means no result is available.
| Benchmark | Qwen3-VL-4B-Instruct | MiniMax M3.1 Flash |
|---|---|---|
| HLE | — | 48.0 |
| DeepSWE | — | 73.7 |
| FrontierCS 2 | — | 3.3 |
| NL2Repo-Bench | — | 61.2 |
| PostTrain Bench | — | 39.3 |
| SWE Refactor Bench | — | 23.0 |
| SWE-sweep | — | 0.67 |
| Terminal-Bench 4.0 | — | 38.1 |
| Vibe Code Bench v1.1 | — | 73.9 |
| Agents' Last Exam v1.0 | — | 29.6 |
| Agents' Last Exam v1.1 | — | 29.5 |
| AutomationBench-AA | — | 72.5 |
| OSWorld 2.0 | — | 61.7 |
| CyberGym | — | 82.2 |
| AA Intelligence Index v4.3 | — | 42.4 |
| AA Intelligence Index v5.0.1 | — | 29.1 |
| Code Arena WebDev | — | 1588 |
| LiveBench 2.0 | — | 73.1 |
| Vals Index v2.2 | — | 53.8 |
| LOCA-Bench 256K | — | 60.3 |
| BabyVision | — | 91.0 |
| CharXiv RQ | — | 82.5 |
| DocVQA | 95.3 | — |
| MMMU | 67.4 | — |
| AutomationBench | — | 62.8 |
| BankerToolBench | — | 76.2 |
| GDPx | — | 4.0 |
| Legal Research Bench | — | 34.1 |
| Tax Agent Bench | — | 32.1 |
Time-of-day providers use peak-hour rates; fixed-price providers use standard rates. Off-peak discounts remain in the published details.
Drag the handle above the bar for cache hits, the one below it for the input : output mix · 5% steps
Blended price = effective input rate at 90% cache hits × 80% + output rate × 20%
A blended cost needs a published input and output price on both sides, and one of these two does not have them yet.
Qwen3-VL-4B-Instruct: Listed text rates; pricing tier not confirmed.
MiniMax M3.1 Flash: Standard pricing. MiniMax international platform (platform.minimax.io) pay-as-you-go list price, flat across context lengths. Official independent currency quote; no conversion.
| Price / 1M tokens | Qwen3-VL-4B-Instruct | MiniMax M3.1 Flash | Ratio |
|---|---|---|---|
| Input | — | $0.15 | — |
| Output | — | $0.60 | — |
| Cached input | — | $0.015 | — |
| Cache write | — | — | — |
| Blended | — | $0.143 | — |
Estimates use the selected token mix. Cache writes, tool fees and long-context surcharges are excluded.
Reported cost per task, paired with the score from the same run. Settings and sources may differ; costs do not affect the verdict.
| Benchmark | Qwen3-VL-4B-Instruct | MiniMax M3.1 Flash |
|---|---|---|
| AA Intelligence Index v5.0.1 | Not reported | |
| Agents' Last Exam v1.0 | Not reported | |
| AutomationBench-AA | Not reported | $0.0665 / task Average · Guardrail-safe objective completion rate: 72.5 Thinking (With Tools) |
| FrontierCS 2 | Not reported | |
| GDPx | Not reported | $1.61 / task Average · GDPx — Score: 4.0 Highest (source-reported) (With Tools) · overall/score |
| HLE | Not reported | |
| Legal Research Bench | Not reported | |
| LiveBench 2.0 | Not reported | $0.1056 / task Average · Overall score (equal-weight mean of 7 suite scores, 0-100): 73.1 Max (task-specific tools) |
| SWE-sweep | Not reported | |
| Tax Agent Bench | Not reported | $0.3209 / task Average · Accuracy (%, rubric × citation quality): 32.1 Unspecified Reasoning (With Tools) |
| Terminal-Bench 4.0 | Not reported | $0.73 / task Average · Resolution rate (%): 25.3 Unspecified Reasoning (With Tools) · vals-mini-swe |
| Vals Index v2.2 | Not reported | |
| Vibe Code Bench v1.1 | Not reported |
| Attribute | Qwen3-VL-4B-Instruct | MiniMax M3.1 Flash |
|---|---|---|
| Availability | Available | Available |
| Provider | Alibaba | MiniMaxAI |
| Model type | Multimodal model | Coding model |
| First release | October 15, 2025 | October 19, 2026 |
| Context window | 256K tokens | 1M tokens |
| Total parameters | 4B | 428B |
| Active parameters | 4B | 23B |
| Architecture | Dense | Mixture of Experts (MoE) |
| Documented capabilities | — | Reasoning, Coding |
| Input modalities | Text, Image, Video | Text, Image, Video |
| Output modalities | Text | Text |
| Weights | Open weights | Open weights |
| Licence | Apache 2.0 | MiniMax M3.1-Flash Community License |
| Commercial use | Commercial use allowed | Proprietary |
These are reported-score comparisons, not controlled head-to-head tests.
What people found comparing these models, or using either one. Comments also appear on the models' own pages.
Used Qwen3-VL-4B-Instruct or MiniMax M3.1 Flash? Sign in with Google to share what worked, what did not, and which API or setup you tested.
No community comments on either model yet.
Alternatives with the most benchmark overlap for this matchup.