Qwen3 VL 235B A22B Thinking vs Kimi K2 Instruct 0905 vs Apertus 70B
Qwen3 VL 235B A22B Thinking comes out ahead, 44 to 38 and 32 on our weighted score, though Kimi K2 Instruct 0905 is 21% cheaper per token.
- Our pick
Alibaba (Qwen)Qwen3 VL 235B A22B Thinking
44/100- ECI—
- Price$0.50 / $3.95
- Context131K
Moonshot AIKimi K2 Instruct 0905
38/100- ECI—
- Price$0.60 / $2.50
- Context262K
Swiss AI
Apertus 70B
32/100- ECI—
- Price$0.82 / $2.42
- Context66K
Qwen3 VL 235B A22B Thinking is our pick
Qwen3 VL 235B A22B Thinking is the better all-round choice, scoring 44/100 against Kimi K2 Instruct 0905 (38) and Apertus 70B (32). It leads on inputs & features. Kimi K2 Instruct 0905 wins on price and context window. The score weighs price 50%, inputs & features 30%, context window 20%. None of the shared benchmarks cover every model here yet, so capability is left out of this verdict.
- CapabilityNot enough dataNo independent benchmark covers every model here yet
- Lowest priceKimi K2 Instruct 0905Kimi K2 Instruct 0905 $1.07 · Apertus 70B $1.22 · Qwen3 VL 235B A22B Thinking $1.36 per 1M tokens (3:1 blend)
- Longest contextKimi K2 Instruct 0905Kimi K2 Instruct 0905 262,144 · Qwen3 VL 235B A22B Thinking 131,072 · Apertus 70B 65,536 tokens
- Widest inputsQwen3 VL 235B A22B ThinkingQwen3 VL 235B A22B Thinking: Text, Images · Kimi K2 Instruct 0905: Text · Apertus 70B: Text
- Self-hostingAll open weightsEvery model here can be downloaded and run on your own hardware
| Measure | Weight | Qwen3 VL 235B A22B Thinking | Kimi K2 Instruct 0905 | Apertus 70B |
|---|---|---|---|---|
| Price | 50% | 42 | 46 | 44 |
| Inputs & features | 30% | 60 | 25 | 25 |
| Context window | 20% | 25 | 37 | 12 |
| Overall | 100% | 44/100 | 38/100 | 32/100 |
Left out because at least one model lacks the data: capability. The remaining weights were rescaled.
Every spec in one table
Highlighted cells lead their row. Dashes mean the data is not published.
| Specification | Qwen3 VL 235B A22B ThinkingAlibaba (Qwen) | Kimi K2 Instruct 0905Moonshot AI | Apertus 70BSwiss AI |
|---|---|---|---|
| Capability | |||
| Capabilities Index (ECI) | — | — | — |
| ECI rank | — | — | — |
| Price per million tokens | |||
| Input | $0.50 (best) | $0.60 | $0.82 |
| Output | $3.95 | $2.50 | $2.42 (best) |
| Cached input | — | — | — |
| Blended (3:1) | $1.36 | $1.07 (best) | $1.22 |
| Long-context rate | Same rate | Same rate | Same rate |
| Price source | Median of 5 providers | Median of 12 providers | Median of 3 providers |
| Limits | |||
| Context window | 131,072 tokens | 262,144 tokens (best) | 65,536 tokens |
| Max output | 32,768 tokens | 98,304 tokens (best) | 8,192 tokens |
| Inputs and features | |||
| Text | Yes | Yes | Yes |
| Images | Yes | No | No |
| PDFs | No | No | No |
| Audio | No | No | No |
| Video | No | No | No |
| Reasoning | Yes | No | No |
| Tool calling | Yes | Yes | Yes |
| Structured output | No | No | No |
| Availability | |||
| Weights | Open | OpenModified MIT | OpenApache-2.0 |
| API model ID | — | — | — |
| API providers | 5 | 13 (best) | 3 |
| Released | Sep 23, 2025 | Sep 5, 2025 | Sep 2, 2025 |
| Knowledge cutoff | Mar 31, 2025 | — | Sep 2025 |
What would a month cost?
Enter your expected volume in millions of tokens. List prices only; caching and batch discounts would lower these.
- Qwen3 VL 235B A22B Thinking$12.90
- Kimi K2 Instruct 0905$11.00
- Apertus 70B$13.04
Which should you choose?
Which is better: Qwen3 VL 235B A22B Thinking, Kimi K2 Instruct 0905 or Apertus 70B?
Qwen3 VL 235B A22B Thinking is the better all-round choice, scoring 44/100 against Kimi K2 Instruct 0905 (38) and Apertus 70B (32). It leads on inputs & features. Kimi K2 Instruct 0905 wins on price and context window. The score weighs price 50%, inputs & features 30%, context window 20%. None of the shared benchmarks cover every model here yet, so capability is left out of this verdict.
Which is cheaper, Qwen3 VL 235B A22B Thinking, Kimi K2 Instruct 0905 or Apertus 70B?
Kimi K2 Instruct 0905 is cheaper at $0.60 input / $2.50 output per million tokens (median across 12 API providers). Apertus 70B costs $0.82 input / $2.42 output per million tokens (median across 3 API providers); Qwen3 VL 235B A22B Thinking costs $0.50 input / $3.95 output per million tokens (median across 5 API providers). At a typical mix of three input tokens to one output token, that is $1.07 per million tokens for Kimi K2 Instruct 0905 versus $1.22 for Apertus 70B (1.1× as much) and $1.36 for Qwen3 VL 235B A22B Thinking (1.3× as much).
Which scores higher on benchmarks?
There is no independent benchmark that covers all three models yet. Qwen3 VL 235B A22B Thinking has not been scored yet, Kimi K2 Instruct 0905 has not been scored yet and Apertus 70B has not been scored yet.
Which is better for coding?
There are no published SWE-bench Verified results for Qwen3 VL 235B A22B Thinking, Kimi K2 Instruct 0905 and Apertus 70B yet, so there is no like-for-like coding score. Test both on a sample of your own tasks. All three support tool calling for agent workflows.
Which has the bigger context window?
Kimi K2 Instruct 0905 has the largest context window at 262,144 tokens, against 131,072 for Qwen3 VL 235B A22B Thinking and 65,536 for Apertus 70B. Maximum output per response: Qwen3 VL 235B A22B Thinking up to 32,768, Kimi K2 Instruct 0905 up to 98,304, Apertus 70B up to 8,192 tokens.
Which can read images, PDFs, audio or video?
Qwen3 VL 235B A22B Thinking accepts text and images; Kimi K2 Instruct 0905 accepts text; Apertus 70B accepts text. Qwen3 VL 235B A22B Thinking handles the widest range of inputs.
Are any of these open source?
Yes, all three publish their weights (Modified MIT and Apache-2.0), so you can self-host them.
Which is newer?
Qwen3 VL 235B A22B Thinking is the newest, released Sep 23, 2025. Kimi K2 Instruct 0905 came out Sep 5, 2025; Apertus 70B came out Sep 2, 2025. Knowledge cutoff: Qwen3 VL 235B A22B Thinking Mar 31, 2025, Apertus 70B Sep 2025.
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.
