Gemma 3 12B IT vs Qwen2.5 7B Instruct vs Llama-3.2-90B-Vision-Instruct
Too close to call on our weighted score (Llama-3.2-90B-Vision-Instruct 44, Gemma 3 12B IT 42, Qwen2.5 7B Instruct 33). The right pick depends on what you value most.
GoogleGemma 3 12B IT
42/100- ECI123.5
- Price$0.05 / $0.15
- Context131K
Alibaba (Qwen)Qwen2.5 7B Instruct
33/100- ECI118.5
- Price$0.05 / $0.10
- Context131K
MetaLlama-3.2-90B-Vision-Instruct
44/100- ECI125.5
- Price—
- Context128K
Too close to call
It is close. Our weighted score puts them within 2 points (Llama-3.2-90B-Vision-Instruct 44/100, Gemma 3 12B IT 42/100, Qwen2.5 7B Instruct 33/100), so choose by what matters most for your work: Llama-3.2-90B-Vision-Instruct for raw capability and Qwen2.5 7B Instruct on price. The score weighs capability 67%, inputs & features 20%, context window 13%.
- CapabilityLlama-3.2-90B-Vision-InstructCapabilities Index (ECI): Llama-3.2-90B-Vision-Instruct 125.5 · Gemma 3 12B IT 123.5 · Qwen2.5 7B Instruct 118.5
- Lowest priceQwen2.5 7B InstructQwen2.5 7B Instruct $0.063 · Gemma 3 12B IT $0.075 per 1M tokens (3:1 blend) · Llama-3.2-90B-Vision-Instruct unpriced
- Longest contextGemma 3 12B IT and Qwen2.5 7B InstructGemma 3 12B IT 131,072 · Qwen2.5 7B Instruct 131,072 · Llama-3.2-90B-Vision-Instruct 128,000 tokens
- Widest inputsGemma 3 12B IT and Llama-3.2-90B-Vision-InstructGemma 3 12B IT: Text, Images · Qwen2.5 7B Instruct: Text · Llama-3.2-90B-Vision-Instruct: Text, Images
- Self-hostingAll open weightsEvery model here can be downloaded and run on your own hardware
| Measure | Weight | Gemma 3 12B IT | Qwen2.5 7B Instruct | Llama-3.2-90B-Vision-Instruct |
|---|---|---|---|---|
| CapabilityCapabilities Index (ECI) | 67% | 44 | 38 | 46 |
| Inputs & features | 20% | 50 | 25 | 50 |
| Context window | 13% | 25 | 25 | 24 |
| Overall | 100% | 42/100 | 33/100 | 44/100 |
Left out because at least one model lacks the data: price. The remaining weights were rescaled.
Every spec in one table
Highlighted cells lead their row. Dashes mean the data is not published.
| Specification | Gemma 3 12B ITGoogle | Qwen2.5 7B InstructAlibaba (Qwen) | Llama-3.2-90B-Vision-InstructMeta |
|---|---|---|---|
| Capability | |||
| Capabilities Index (ECI) | 123.5 | 118.5 | 125.5 (best) |
| ECI rank | #151 of 165 | #159 of 165 | #150 of 165 (best) |
| GPQA DiamondGraduate-level science questions | 39.5% | 35.5% | 41.0% (best) |
| OTIS Mock AIME 2024–2025Competition mathematics | 16.7% (best) | 2.5% | 2.6% |
| Price per million tokens | |||
| Input | $0.05 | $0.05 | — |
| Output | $0.15 | $0.10 (best) | — |
| Cached input | — | — | — |
| Blended (3:1) | $0.075 | $0.063 (best) | — |
| Long-context rate | Same rate | Same rate | — |
| Price source | Median of 12 providers | Median of 3 providers | — |
| Limits | |||
| Context window | 131,072 tokens (best) | 131,072 tokens (best) | 128,000 tokens |
| Max output | 131,072 tokens (best) | 8,192 tokens | 4,096 tokens |
| Inputs and features | |||
| Text | Yes | Yes | Yes |
| Images | Yes | No | Yes |
| PDFs | No | No | No |
| Audio | No | No | No |
| Video | No | No | No |
| Reasoning | No | No | No |
| Tool calling | Yes | Yes | Yes |
| Structured output | No | No | No |
| Availability | |||
| Weights | Open | OpenApache 2.0 | OpenLlama 3.2 Community License |
| API model ID | — | — | — |
| API providers | 13 (best) | 4 | 1 |
| Released | Mar 12, 2025 | Sep 19, 2024 | Sep 25, 2024 |
| Knowledge cutoff | Aug 2024 | — | Dec 2023 |
What would a month cost?
Enter your expected volume in millions of tokens. List prices only; caching and batch discounts would lower these.
- Gemma 3 12B IT$0.80
- Qwen2.5 7B Instruct$0.70
- Llama-3.2-90B-Vision-Instruct—
Which should you choose?
Which is better: Gemma 3 12B IT, Qwen2.5 7B Instruct or Llama-3.2-90B-Vision-Instruct?
It is close. Our weighted score puts them within 2 points (Llama-3.2-90B-Vision-Instruct 44/100, Gemma 3 12B IT 42/100, Qwen2.5 7B Instruct 33/100), so choose by what matters most for your work: Llama-3.2-90B-Vision-Instruct for raw capability and Qwen2.5 7B Instruct on price. The score weighs capability 67%, inputs & features 20%, context window 13%.
Which is cheaper, Gemma 3 12B IT, Qwen2.5 7B Instruct or Llama-3.2-90B-Vision-Instruct?
Qwen2.5 7B Instruct is cheaper at $0.05 input / $0.10 output per million tokens (median across 3 API providers). Gemma 3 12B IT costs $0.05 input / $0.15 output per million tokens (median across 12 API providers). At a typical mix of three input tokens to one output token, that is $0.063 per million tokens for Qwen2.5 7B Instruct versus $0.075 for Gemma 3 12B IT (1.2× as much). Llama-3.2-90B-Vision-Instruct has no published per-token price.
Which scores higher on benchmarks?
Llama-3.2-90B-Vision-Instruct scores higher on the Capabilities Index (ECI): Llama-3.2-90B-Vision-Instruct 125.5 (#150 of 165), Gemma 3 12B IT 123.5 (#151 of 165) and Qwen2.5 7B Instruct 118.5 (#159 of 165). The confidence ranges of the top two overlap (120.3–127.4 vs 116.9–128.4), so treat the gap as small. On individual benchmarks: GPQA Diamond — Llama-3.2-90B-Vision-Instruct 41.0%, Gemma 3 12B IT 39.5%, Qwen2.5 7B Instruct 35.5%; OTIS Mock AIME 2024–2025 — Gemma 3 12B IT 16.7%, Llama-3.2-90B-Vision-Instruct 2.6%, Qwen2.5 7B Instruct 2.5%.
Which is better for coding?
There are no published SWE-bench Verified results for Gemma 3 12B IT, Qwen2.5 7B Instruct and Llama-3.2-90B-Vision-Instruct yet, so there is no like-for-like coding score. On overall capability, Llama-3.2-90B-Vision-Instruct leads, which tends to carry over to coding, but test on your own codebase. All three support tool calling for agent workflows.
Which has the bigger context window?
Gemma 3 12B IT and Qwen2.5 7B Instruct have the largest context windows (131,072 and 131,072 tokens), against 128,000 for Llama-3.2-90B-Vision-Instruct. Maximum output per response: Gemma 3 12B IT up to 131,072, Qwen2.5 7B Instruct up to 8,192, Llama-3.2-90B-Vision-Instruct up to 4,096 tokens.
Which can read images, PDFs, audio or video?
Gemma 3 12B IT accepts text and images; Qwen2.5 7B Instruct accepts text; Llama-3.2-90B-Vision-Instruct accepts text and images. Gemma 3 12B IT handles the widest range of inputs.
Are any of these open source?
Yes, all three publish their weights (Apache 2.0 and Llama 3.2 Community License), so you can self-host them.
Which is newer?
Gemma 3 12B IT is the newest, released Mar 12, 2025. Llama-3.2-90B-Vision-Instruct came out Sep 25, 2024; Qwen2.5 7B Instruct came out Sep 19, 2024. Knowledge cutoff: Gemma 3 12B IT Aug 2024, Llama-3.2-90B-Vision-Instruct Dec 2023.
How do you decide the winner?
Each model gets a 0–100 score on capability (50%, independent benchmark results); price (25%, blended price per million tokens (3 input : 1 output), log scale); inputs & features (15%, image, PDF, audio and video input, tool calling, structured output and reasoning); context window (10%, maximum tokens per request, log scale). Dimensions missing for any model are dropped and the remaining weights rescaled, so every model is judged on the same evidence. Specs and prices come from public model listings and the labs’ own API pages; capability scores come from independent benchmark runs. Data updated Nov 2, 2026.
Related comparisons
- Gemma 3 12B IT vs Mistral NemoMistral AI · ECI 118.6
- Gemma 3 12B IT vs GPT-4o miniOpenAI · ECI 126.5
- Qwen2.5 7B Instruct vs Llama-3.1-8B-InstructMeta · ECI 116.4
- Qwen2.5 7B Instruct vs Mistral NemoMistral AI · ECI 118.6
- Qwen2.5 7B Instruct vs Gemma 3 4B ITGoogle · ECI 115.8
- Qwen2.5 7B Instruct vs Claude Haiku 3Anthropic · ECI 118.8
