DeepSeek-V3 vs QwQ 32B vs Llama 4 Maverick 17B Instruct
Llama 4 Maverick 17B Instruct comes out ahead, 58 to 53 and 50 on our weighted score, and it is the cheaper option too.
DeepSeekDeepSeek-V3
50/100- ECI132.3
- Price$0.32 / $1.10
- Context131K
Alibaba (Qwen)QwQ 32B
53/100- ECI137.6
- Price$0.66 / $1.00
- Context131K
- Our pick
MetaLlama 4 Maverick 17B Instruct
58/100- ECI132.2
- Price$0.321 / $0.91
- Context1M
Llama 4 Maverick 17B Instruct is our pick
Llama 4 Maverick 17B Instruct is the better all-round choice, scoring 58/100 against QwQ 32B (53) and DeepSeek-V3 (50). It leads on price, inputs & features and context window. QwQ 32B wins on capability. The score weighs capability 50%, price 25%, inputs & features 15%, context window 10%.
- CapabilityQwQ 32BCapabilities Index (ECI): QwQ 32B 137.6 · DeepSeek-V3 132.3 · Llama 4 Maverick 17B Instruct 132.2
- Lowest priceLlama 4 Maverick 17B InstructLlama 4 Maverick 17B Instruct $0.468 · DeepSeek-V3 $0.515 · QwQ 32B $0.745 per 1M tokens (3:1 blend)
- Longest contextLlama 4 Maverick 17B InstructLlama 4 Maverick 17B Instruct 1,000,000 · DeepSeek-V3 131,072 · QwQ 32B 131,072 tokens
- Widest inputsLlama 4 Maverick 17B InstructDeepSeek-V3: Text · QwQ 32B: Text · Llama 4 Maverick 17B Instruct: Text, Images
- Self-hostingAll open weightsEvery model here can be downloaded and run on your own hardware
| Measure | Weight | DeepSeek-V3 | QwQ 32B | Llama 4 Maverick 17B Instruct |
|---|---|---|---|---|
| CapabilityCapabilities Index (ECI) | 50% | 56 | 62 | 56 |
| Price | 25% | 64 | 56 | 66 |
| Inputs & features | 15% | 25 | 35 | 50 |
| Context window | 10% | 24 | 24 | 60 |
| Overall | 100% | 50/100 | 53/100 | 58/100 |
Every spec in one table
Highlighted cells lead their row. Dashes mean the data is not published.
| Specification | DeepSeek-V3DeepSeek | QwQ 32BAlibaba (Qwen) | Llama 4 Maverick 17B InstructMeta |
|---|---|---|---|
| Capability | |||
| Capabilities Index (ECI) | 132.3 | 137.6 (best) | 132.2 |
| ECI rank | #121 of 148 | #109 of 148 (best) | #122 of 148 |
| GPQA DiamondGraduate-level science questions | 56.5% | 65.3% | 67.0% (best) |
| OTIS Mock AIME 2024–2025Competition mathematics | 15.8% | 59.2% (best) | 20.6% |
| Price per million tokens | |||
| Input | $0.32 (best) | $0.66 | $0.321 |
| Output | $1.10 | $1.00 | $0.91 (best) |
| Cached input | — | — | — |
| Blended (3:1) | $0.515 | $0.745 | $0.468 (best) |
| Long-context rate | Same rate | Same rate | Same rate |
| Price source | Median of 5 providers | Median of 1 providers | Median of 6 providers |
| Limits | |||
| Context window | 131,072 tokens | 131,072 tokens | 1,000,000 tokens (best) |
| Max output | 8,192 tokens | 8,192 tokens | 16,384 tokens (best) |
| Inputs and features | |||
| Text | Yes | Yes | Yes |
| Images | No | No | Yes |
| PDFs | No | No | No |
| Audio | No | No | No |
| Video | No | No | No |
| Reasoning | No | Yes | No |
| Tool calling | Yes | Yes | Yes |
| Structured output | No | No | No |
| Availability | |||
| Weights | OpenDeepSeek Model License | Open | Open |
| API model ID | — | — | — |
| API providers | 5 | 1 | 6 (best) |
| Released | Dec 26, 2024 | Mar 5, 2025 | Apr 5, 2025 |
| Knowledge cutoff | — | Apr 2024 | Aug 2024 |
What would a month cost?
Enter your expected volume in millions of tokens. List prices only; caching and batch discounts would lower these.
- DeepSeek-V3$5.40
- QwQ 32B$8.60
- Llama 4 Maverick 17B Instruct$5.03
Which should you choose?
Which is better: DeepSeek-V3, QwQ 32B or Llama 4 Maverick 17B Instruct?
Llama 4 Maverick 17B Instruct is the better all-round choice, scoring 58/100 against QwQ 32B (53) and DeepSeek-V3 (50). It leads on price, inputs & features and context window. QwQ 32B wins on capability. The score weighs capability 50%, price 25%, inputs & features 15%, context window 10%.
Which is cheaper, DeepSeek-V3, QwQ 32B or Llama 4 Maverick 17B Instruct?
Llama 4 Maverick 17B Instruct is cheaper at $0.321 input / $0.91 output per million tokens (median across 6 API providers). DeepSeek-V3 costs $0.32 input / $1.10 output per million tokens (median across 5 API providers); QwQ 32B costs $0.66 input / $1.00 output per million tokens (median across 1 API provider). At a typical mix of three input tokens to one output token, that is $0.468 per million tokens for Llama 4 Maverick 17B Instruct versus $0.515 for DeepSeek-V3 (1.1× as much) and $0.745 for QwQ 32B (1.6× as much).
Which scores higher on benchmarks?
QwQ 32B scores higher on the Capabilities Index (ECI): QwQ 32B 137.6 (#109 of 148), DeepSeek-V3 132.3 (#121 of 148) and Llama 4 Maverick 17B Instruct 132.2 (#122 of 148). The confidence ranges of the top two overlap (133.1–141.7 vs 127.5–135.5), so treat the gap as small. On individual benchmarks: GPQA Diamond — Llama 4 Maverick 17B Instruct 67.0%, QwQ 32B 65.3%, DeepSeek-V3 56.5%; OTIS Mock AIME 2024–2025 — QwQ 32B 59.2%, Llama 4 Maverick 17B Instruct 20.6%, DeepSeek-V3 15.8%.
Which is better for coding?
There are no published SWE-bench Verified results for DeepSeek-V3, QwQ 32B and Llama 4 Maverick 17B Instruct yet, so there is no like-for-like coding score. On overall capability, QwQ 32B 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?
Llama 4 Maverick 17B Instruct has the largest context window at 1,000,000 tokens, against 131,072 for DeepSeek-V3 and 131,072 for QwQ 32B. Maximum output per response: DeepSeek-V3 up to 8,192, QwQ 32B up to 8,192, Llama 4 Maverick 17B Instruct up to 16,384 tokens.
Which can read images, PDFs, audio or video?
DeepSeek-V3 accepts text; QwQ 32B accepts text; Llama 4 Maverick 17B Instruct accepts text and images. Llama 4 Maverick 17B Instruct handles the widest range of inputs.
Are any of these open source?
Yes, all three publish their weights (DeepSeek Model License), so you can self-host them.
Which is newer?
Llama 4 Maverick 17B Instruct is the newest, released Apr 5, 2025. QwQ 32B came out Mar 5, 2025; DeepSeek-V3 came out Dec 26, 2024. Knowledge cutoff: QwQ 32B Apr 2024, Llama 4 Maverick 17B Instruct Aug 2024.
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 Oct 4, 2026.
