Kimi K2.6 vs Kimi K2 Instruct vs Qwen3 235B-A22B Thinking 2507
Kimi K2.6 comes out ahead, 65 to 57 and 53 on our weighted score, though Qwen3 235B-A22B Thinking 2507 is 45% cheaper per token.
Price in rating: Included · Exclude price
- Our pick
Moonshot AI
Kimi K2.6
65/100- ECI151.6
- Price$0.95 / $4.00
- Context262K
Moonshot AI
Kimi K2 Instruct
53/100- ECI141.3
- Price$0.57 / $2.30
- Context131K
Alibaba (Qwen)
Qwen3 235B-A22B Thinking 2507
57/100- ECI143.3
- Price$0.30 / $2.90
- Context262K
Kimi K2.6 is our pick
Kimi K2.6 is the better all-round choice, scoring 65/100 against Qwen3 235B-A22B Thinking 2507 (57) and Kimi K2 Instruct (53). It leads on capability and inputs & features. The score weighs capability 50%, price 25%, inputs & features 15%, context window 10%.
- CapabilityKimi K2.6Capabilities Index (ECI): Kimi K2.6 151.6 · Qwen3 235B-A22B Thinking 2507 143.3 · Kimi K2 Instruct 141.3
- Lowest priceQwen3 235B-A22B Thinking 2507Qwen3 235B-A22B Thinking 2507 $0.95 · Kimi K2 Instruct $1.00 · Kimi K2.6 $1.71 per 1M tokens (3:1 blend)
- Longest contextKimi K2.6 and Qwen3 235B-A22B Thinking 2507Kimi K2.6 262,144 · Qwen3 235B-A22B Thinking 2507 262,144 · Kimi K2 Instruct 131,072 tokens
- Widest inputsKimi K2.6Kimi K2.6: Text, Images, Video · Kimi K2 Instruct: Text · Qwen3 235B-A22B Thinking 2507: Text
- Self-hostingAll open weightsEvery model here can be downloaded and run on your own hardware
| Measure | Weight | Kimi K2.6 | Kimi K2 Instruct | Qwen3 235B-A22B Thinking 2507 |
|---|---|---|---|---|
| CapabilityCapabilities Index (ECI) | 50% | 79 | 69 | 71 |
| Price | 25% | 37 | 48 | 49 |
| Inputs & features | 15% | 80 | 25 | 35 |
| Context window | 10% | 43 | 32 | 43 |
| Overall | 100% | 65/100 | 53/100 | 57/100 |
Every spec in one table
Highlighted cells lead their row. Dashes mean the data is not published.
| Specification | |||
|---|---|---|---|
| Capability | |||
| Capabilities Index (ECI) | 151.6 (best) | 141.3 | 143.3 |
| ECI rank | #61 of 158 (best) | #108 of 158 | #100 of 158 |
| GPQA DiamondGraduate-level science questions | 90.8% (best) | — | 80.1% |
| FrontierMath Tiers 1–3Research-level mathematics | 57.2% | — | — |
| OTIS Mock AIME 2024–2025Competition mathematics | 96.1% (best) | — | 86.7% |
| SWE-bench VerifiedFixing real GitHub issues | 76.7% | — | — |
| SimpleQA VerifiedShort factual questions | 34.9% | — | 40.4% (best) |
| Price per million tokens | |||
| Input | $0.95 | $0.57 | $0.30 (best) |
| Output | $4.00 | $2.30 (best) | $2.90 |
| Cached input | $0.16 | — | — |
| Blended (3:1) | $1.71 | $1.00 | $0.95 (best) |
| Long-context rate | Same rate | Same rate | Same rate |
| Price source | Official Moonshot AI API | Median of 11 providers | Median of 11 providers |
| Limits | |||
| Context window | 262,144 tokens (best) | 131,072 tokens | 262,144 tokens (best) |
| Max output | 262,144 tokens (best) | 32,768 tokens | 81,920 tokens |
| Inputs and features | |||
| Text | Yes | Yes | Yes |
| Images | Yes | No | No |
| PDFs | No | No | No |
| Audio | No | No | No |
| Video | Yes | No | No |
| Reasoning | Yes | No | Yes |
| Tool calling | Yes | Yes | Yes |
| Structured output | Yes | No | No |
| Availability | |||
| Weights | Open | OpenModified MIT | OpenApache 2.0 |
| API model ID | kimi-k2.6 | — | — |
| API providers | 64 (best) | 11 | 12 |
| Released | Apr 20, 2026 | Jul 11, 2025 | Jul 25, 2025 |
| Knowledge cutoff | Jan 2025 | — | — |
What would a month cost?
Enter your expected volume in millions of tokens. List prices only; caching and batch discounts would lower these.
Kimi K2.6$17.50
Kimi K2 Instruct$10.30
Qwen3 235B-A22B Thinking 2507$8.80
Which should you choose?
Which is better: Kimi K2.6, Kimi K2 Instruct or Qwen3 235B-A22B Thinking 2507?
Kimi K2.6 is the better all-round choice, scoring 65/100 against Qwen3 235B-A22B Thinking 2507 (57) and Kimi K2 Instruct (53). It leads on capability and inputs & features. The score weighs capability 50%, price 25%, inputs & features 15%, context window 10%.
Which is cheaper, Kimi K2.6, Kimi K2 Instruct or Qwen3 235B-A22B Thinking 2507?
Qwen3 235B-A22B Thinking 2507 is cheaper at $0.30 input / $2.90 output per million tokens (median across 11 API providers). Kimi K2 Instruct costs $0.57 input / $2.30 output per million tokens (median across 11 API providers); Kimi K2.6 costs $0.95 input / $4.00 output per million tokens (official Moonshot AI API price). At a typical mix of three input tokens to one output token, that is $0.95 per million tokens for Qwen3 235B-A22B Thinking 2507 versus $1.00 for Kimi K2 Instruct (1.1× as much) and $1.71 for Kimi K2.6 (1.8× as much).
Which scores higher on benchmarks?
Kimi K2.6 scores higher on the Capabilities Index (ECI): Kimi K2.6 151.6 (#61 of 158), Qwen3 235B-A22B Thinking 2507 143.3 (#100 of 158) and Kimi K2 Instruct 141.3 (#108 of 158). Their confidence ranges do not overlap (149.9–153.4 vs 141.4–145.1), so the gap is a real one.
Which is better for coding?
There are no published SWE-bench Verified results for Kimi K2 Instruct and Qwen3 235B-A22B Thinking 2507 yet, so there is no like-for-like coding score. On overall capability, Kimi K2.6 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?
Kimi K2.6 and Qwen3 235B-A22B Thinking 2507 have the largest context windows (262,144 and 262,144 tokens), against 131,072 for Kimi K2 Instruct. Maximum output per response: Kimi K2.6 up to 262,144, Kimi K2 Instruct up to 32,768, Qwen3 235B-A22B Thinking 2507 up to 81,920 tokens.
Which can read images, PDFs, audio or video?
Kimi K2.6 accepts text, images and video; Kimi K2 Instruct accepts text; Qwen3 235B-A22B Thinking 2507 accepts text. Kimi K2.6 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?
Kimi K2.6 is the newest, released Apr 20, 2026. Qwen3 235B-A22B Thinking 2507 came out Jul 25, 2025; Kimi K2 Instruct came out Jul 11, 2025. Knowledge cutoff: Kimi K2.6 Jan 2025.
How do you decide the winner?
This rating uses 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 18, 2026.