Skip to content

MBZUAI Institute of Foundation Models · Model profile

K2 Think V2

Intelligence, Performance & Price Analysis

Source: Canonical slug: k2-think-v2 · Canonical model registry at build time

Routing verdict

Capable everyday model

Good for routine tasks; route complex reasoning and premium workloads to stronger models.

K2 Think V2 summary metrics

Intelligence
17 scoreK2 Think V2 scores 17 on the Artificial Analysis Intelligence Index, placing it well above average among other open weight models of similar size (median: 9).
Input Price
Not published
Output Price
Not published
Verbosity
110M Output tokens from Intelligence Index 4 out of 4 units for Verbosity . Comp

Assessment

Routing verdict and task fit

K2 Think V2 scores 17 on the Artificial Analysis Intelligence Index, placing it well above average among other open weight models of similar size (median: 9).
Task fit assessment
WorkloadRatingNotes
Complex reasoning & agentic workflowsViableIntelligence score 17 handles routine reasoning but may struggle with open-ended agentic tasks.

Evidence

Evidence charts

13 published comparison charts for this profile. K2 Think V2 is marked with a diamond and bold label; each chart has a table view and lists its source and retrieval date.

AA-Omniscience Index

AA-Omniscience Index (higher is better) measures knowledge reliability and hallucination. It rewards correct answers, penalizes hallucinations, and has no penalty for refusing to answer. Scores range from -100 to 100, where 0 means as many correct as incorrect answers, and negative scores mean more incorrect than correct. · Evaluation results measured independently by Artificial Analysis

20 models in the published order. K2 Think V2: -33.9, row 12 of 20 in the published order.
  1. Claude Fable 5 (with fallback)40.1
  2. Gemini 3.5 Flash22.7
  3. GPT-5.5 (xhigh)20.1
  4. Grok 4.3 (high)18.3
  5. Kimi K2.66.4
  6. Muse Spark4.1
  7. GLM-5.2 (max)4
  8. MiniMax-M31.4
  9. Nemotron 3 Ultra-0.833
  10. DeepSeek V4 Pro (max)-10
  11. Mistral Small 4-29.9
  12. K2 Think V2 (this model)-33.9
  13. Mistral Medium 3.5-36.3
  14. Qwen3.5 122B A10B-39.6
  15. NVIDIA Nemotron 3 Super-42.1
  16. Devstral 2-46.2
  17. gpt-oss-120b (high)-50
  18. Qwen3 Next 80B A3B-50.6
  19. Llama 3.3 70B-51.9
  20. Llama 4 Scout-52.4

Published benchmark dataset · Retrieved

Method and source
Methodology
Independent test run by Artificial Analysis on dedicated hardware.
Chart interpretation
Each bar is one model measured the same way, so this model can be read against the others. The chart lists its source and retrieval date.

Artificial Analysis Intelligence Index by Open Weights / Proprietary

Artificial Analysis Intelligence Index v4.1 incorporates 9 evaluations: GDPval-AA v2, 𝜏³-Banking, Terminal-Bench v2.1, SciCode, Humanity's Last Exam, GPQA Diamond, CritPt, AA-Omniscience, AA-LCR · Evaluation results measured independently by Artificial Analysis

20 models in the published order. K2 Think V2: 17.3, row 19 of 20 in the published order.
  1. Claude Fable 5 (with fallback)59.9
  2. GPT-5.5 (xhigh)54.8
  3. GLM-5.2 (max)51.1
  4. Gemini 3.5 Flash50.2
  5. MiniMax-M344.4
  6. DeepSeek V4 Pro (max)44.3
  7. Kimi K2.644.2
  8. Muse Spark43.1
  9. Nemotron 3 Ultra37.8
  10. Grok 4.3 (high)37.6
  11. Qwen3.5 122B A10B32.3
  12. Mistral Medium 3.529.9
  13. NVIDIA Nemotron 3 Super25.4
  14. gpt-oss-120b (high)23.8
  15. Qwen3 Coder Next21.1
  16. Mistral Small 419.6
  17. Devstral 219.2
  18. HyperNova 60B 260517.8
  19. K2 Think V2 (this model)17.3
  20. Qwen3 Next 80B A3B16.7

Published benchmark dataset · Retrieved

Method and source
Methodology
Independent test run by Artificial Analysis on dedicated hardware.
Benchmark interpretation
Higher values indicate stronger overall capability. Bars show how models compare on composite intelligence metrics.

Artificial Analysis Intelligence Index

Artificial Analysis Intelligence Index v4.1 incorporates 9 evaluations: GDPval-AA v2, 𝜏³-Banking, Terminal-Bench v2.1, SciCode, Humanity's Last Exam, GPQA Diamond, CritPt, AA-Omniscience, AA-LCR · Evaluation results measured independently by Artificial Analysis

20 models in the published order. K2 Think V2: 17.3, row 19 of 20 in the published order.
  1. Claude Fable 5 (with fallback)59.9
  2. GPT-5.5 (xhigh)54.8
  3. GLM-5.2 (max)51.1
  4. Gemini 3.5 Flash50.2
  5. MiniMax-M344.4
  6. DeepSeek V4 Pro (max)44.3
  7. Kimi K2.644.2
  8. Muse Spark43.1
  9. Nemotron 3 Ultra37.8
  10. Grok 4.3 (high)37.6
  11. Qwen3.5 122B A10B32.3
  12. Mistral Medium 3.529.9
  13. NVIDIA Nemotron 3 Super25.4
  14. gpt-oss-120b (high)23.8
  15. Qwen3 Coder Next21.1
  16. Mistral Small 419.6
  17. Devstral 219.2
  18. HyperNova 60B 260517.8
  19. K2 Think V2 (this model)17.3
  20. Qwen3 Next 80B A3B16.7

Published benchmark dataset · Retrieved

Method and source
Methodology
Independent test run by Artificial Analysis on dedicated hardware.
Benchmark interpretation
Higher values indicate stronger overall capability. Bars show how models compare on composite intelligence metrics.

Artificial Analysis Openness Index: Score

Openness Index assesses model openness on a 0 to 100 normalized scale (higher is more open) · Evaluation results measured independently by Artificial Analysis

17 models in the published order. K2 Think V2: 88.9, row 1 of 17 in the published order.
  1. K2 Think V2 (this model)88.9
  2. Nemotron 3 Ultra83.3
  3. NVIDIA Nemotron 3 Super83.3
  4. DeepSeek V4 Pro (max)50
  5. GLM-5.2 (max)44.4
  6. Qwen3 Next 80B A3B44.4
  7. Qwen3 Coder Next41.7
  8. gpt-oss-120b (high)38.9
  9. Qwen3.5 122B A10B38.9
  10. Mistral Small 438.9
  11. HyperNova 60B 260538.9
  12. Ling 2.6 Flash38.9
  13. Llama 3.3 70B38.9
  14. Kimi K2.633.3
  15. MiniMax-M333.3
  16. Mistral Medium 3.533.3
  17. Llama 4 Scout27.8

Published benchmark dataset · Retrieved

Method and source
Methodology
Independent test run by Artificial Analysis on dedicated hardware.
Chart interpretation
Each bar is one model measured the same way, so this model can be read against the others. The chart lists its source and retrieval date.

Intelligence

Artificial Analysis Intelligence Index · Higher is better · Evaluation results measured independently by Artificial Analysis

12 models in the published order. K2 Think V2: 17.3, row 12 of 12 in the published order.
  1. Claude Fable 5 (with fallback)59.9
  2. GPT-5.5 (xhigh)54.8
  3. GLM-5.2 (max)51.1
  4. Gemini 3.5 Flash50.2
  5. MiniMax-M344.4
  6. DeepSeek V4 Pro (max)44.3
  7. Kimi K2.644.2
  8. Muse Spark43.1
  9. Nemotron 3 Ultra37.8
  10. Grok 4.3 (high)37.6
  11. gpt-oss-120b (high)23.8
  12. K2 Think V2 (this model)17.3

Published benchmark dataset · Retrieved

Method and source
Methodology
Independent test run by Artificial Analysis on dedicated hardware.
Benchmark interpretation
Higher values indicate stronger overall capability. Bars show how models compare on composite intelligence metrics.

Speed

Output tokens per second · Higher is better · Evaluation results measured independently by Artificial Analysis

10 models in the published order. K2 Think V2 is not marked in this published set of 10 models.
  1. gpt-oss-120b (high)314 t/s
  2. Nemotron 3 Ultra250 t/s
  3. GLM-5.2 (max)218 t/s
  4. Gemini 3.5 Flash192 t/s
  5. Grok 4.3 (high)164 t/s
  6. MiniMax-M399 t/s
  7. GPT-5.5 (xhigh)88 t/s
  8. Kimi K2.676 t/s
  9. DeepSeek V4 Pro (max)72 t/s
  10. Claude Fable 5 (with fallback)71 t/s

Published benchmark dataset · Retrieved

Method and source
Methodology
Independent test run by Artificial Analysis on dedicated hardware.
Output speed context
Speed should be read together with latency and cost; fast output does not always mean better user-perceived performance.

Context Window

Context window: tokens limit · Higher is better · Evaluation results measured independently by Artificial Analysis

20 models in the published order. K2 Think V2: 262.1k, row 10 of 20 in the published order.
  1. Llama 4 Scout10M
  2. Gemini 3.5 Flash1M
  3. Claude Fable 5 (with fallback)1M
  4. GLM-5.2 (max)1M
  5. DeepSeek V4 Pro (max)1M
  6. Grok 4.3 (high)1M
  7. MiniMax-M31M
  8. NVIDIA Nemotron 3 Super1M
  9. GPT-5.5 (xhigh)922k
  10. K2 Think V2 (this model)262.1k
  11. Nemotron 3 Ultra262.1k
  12. Muse Spark262.1k
  13. Qwen3.5 122B A10B262.1k
  14. Qwen3 Next 80B A3B262.1k
  15. Ling 2.6 Flash262.1k
  16. Kimi K2.6256k
  17. Mistral Medium 3.5256k
  18. Qwen3 Coder Next256k
  19. Mistral Small 4256k
  20. Devstral 2256k

Published benchmark dataset · Retrieved

Method and source
Methodology
Independent test run by Artificial Analysis on dedicated hardware.
Context window sizing
Larger context windows enable complex multi-turn tasks but may increase latency and cost for long inputs.

Cost per Intelligence Index Task

Weighted average cost (USD) per Artificial Analysis Intelligence Index task, segmented by token type. Lower is better · Evaluation results measured independently by Artificial Analysis

18 models in the published order. K2 Think V2 is not marked in this published set of 18 models.
  1. Llama 4 Scout0.000824
  2. HyperNova 60B 26050.000717
  3. DeepSeek V4 Pro (max)0.0063
  4. gpt-oss-120b (high)0.00454
  5. Llama 3.3 70B0.0012
  6. MiniMax-M30.0139
  7. Grok 4.3 (high)0.0101
  8. Qwen3 Next 80B A3B0.0111
  9. NVIDIA Nemotron 3 Super0.00577
  10. Qwen3.5 122B A10B0.0208
  11. Nemotron 3 Ultra0.0209
  12. Qwen3 Coder Next0.0184
  13. Kimi K2.60.0216
  14. GLM-5.2 (max)0.0268
  15. Gemini 3.5 Flash0.0884
  16. GPT-5.5 (xhigh)0.107
  17. Mistral Medium 3.50.0529
  18. Claude Fable 5 (with fallback)0.367

Published benchmark dataset · Retrieved

Method and source
Methodology
Independent test run by Artificial Analysis on dedicated hardware.
Benchmark interpretation
Higher values indicate stronger overall capability. Bars show how models compare on composite intelligence metrics.
Cost pressure
Ethen Gateway should route away from this model when the task is simple, repetitive, or likely to generate many output tokens.

Cost per Task

Weighted average cost (USD) per Intelligence Index task · Lower is better · Evaluation results measured independently by Artificial Analysis

10 models in the published order. K2 Think V2 is not marked in this published set of 10 models.
  1. DeepSeek V4 Pro (max)$0.045
  2. gpt-oss-120b (high)$0.061
  3. MiniMax-M3$0.125
  4. Grok 4.3 (high)$0.139
  5. Nemotron 3 Ultra$0.244
  6. Kimi K2.6$0.346
  7. GLM-5.2 (max)$0.370
  8. Gemini 3.5 Flash$0.586
  9. GPT-5.5 (xhigh)$0.863
  10. Claude Fable 5 (with fallback)$2.75

Published benchmark dataset · Retrieved

Method and source
Methodology
Independent test run by Artificial Analysis on dedicated hardware.
Cost pressure
Ethen Gateway should route away from this model when the task is simple, repetitive, or likely to generate many output tokens.

Cost to Run Artificial Analysis Intelligence Index

Cost (USD) to run all evaluations in the Artificial Analysis Intelligence Index · Evaluation results measured independently by Artificial Analysis

18 models in the published order. K2 Think V2 is not marked in this published set of 18 models.
  1. Llama 4 Scout2.7
  2. HyperNova 60B 26050.947
  3. Llama 3.3 70B2.7
  4. gpt-oss-120b (high)3.5
  5. DeepSeek V4 Pro (max)7.2
  6. MiniMax-M314.5
  7. Qwen3 Coder Next38.5
  8. NVIDIA Nemotron 3 Super5.4
  9. Grok 4.3 (high)12.5
  10. Qwen3 Next 80B A3B15.9
  11. Nemotron 3 Ultra33.5
  12. Qwen3.5 122B A10B32.7
  13. GLM-5.2 (max)32.9
  14. Kimi K2.625
  15. Gemini 3.5 Flash114
  16. Mistral Medium 3.553.7
  17. GPT-5.5 (xhigh)131.8
  18. Claude Fable 5 (with fallback)507.6

Published benchmark dataset · Retrieved

Method and source
Methodology
Independent test run by Artificial Analysis on dedicated hardware.
Benchmark interpretation
Higher values indicate stronger overall capability. Bars show how models compare on composite intelligence metrics.
Cost pressure
Ethen Gateway should route away from this model when the task is simple, repetitive, or likely to generate many output tokens.

Pricing: Cache Hit, Input, and Output

Price (USD per M Tokens) · Evaluation results measured independently by Artificial Analysis

20 models; segments: Input, Output, Cache Hit. K2 Think V2 is not marked in this published set of 20 models.
  • Input
  • Output
  • Cache Hit
  1. HyperNova 60B 2605$0.180
  2. Ling 2.6 Flash$0.420
  3. gpt-oss-120b (high)$0.900
  4. Mistral Small 4$0.750
  5. Llama 4 Scout$0.830
  6. NVIDIA Nemotron 3 Super$1.17
  7. Llama 3.3 70B$1.87
  8. DeepSeek V4 Pro (max)$1.31
  9. MiniMax-M3$1.56
  10. Qwen3 Coder Next$1.90
  11. Nemotron 3 Ultra$3.60
  12. Qwen3.5 122B A10B$3.60
  13. Grok 4.3 (high)$3.95
  14. Kimi K2.6$5.11
  15. GLM-5.2 (max)$6.06
  16. Qwen3 Next 80B A3B$6.50
  17. Mistral Medium 3.5$9.15
  18. Gemini 3.5 Flash$10.65
  19. GPT-5.5 (xhigh)$35.50
  20. Claude Fable 5 (with fallback)$61.00

Published benchmark dataset · Retrieved

Method and source
Methodology
Independent test run by Artificial Analysis on dedicated hardware.
Chart interpretation
Each bar is one model measured the same way, so this model can be read against the others. The chart lists its source and retrieval date.

Model Size: Total and Active Parameters

Comparison between total model parameters and parameters active during inference · Evaluation results measured independently by Artificial Analysis

18 models; segments: Passive Params, Active Params. K2 Think V2: 70B, row 16 of 18 in the published order.
  • Passive Params
  • Active Params
  1. DeepSeek V4 Pro (max)1,600B
  2. Kimi K2.61,000B
  3. GLM-5.2 (max)753B
  4. Nemotron 3 Ultra550B
  5. MiniMax-M3428B
  6. Mistral Medium 3.5128B
  7. Qwen3.5 122B A10B125B
  8. Devstral 2125B
  9. NVIDIA Nemotron 3 Super120.6B
  10. Mistral Small 4119B
  11. gpt-oss-120b (high)117B
  12. Llama 4 Scout109B
  13. Ling 2.6 Flash107B
  14. Qwen3 Next 80B A3B80B
  15. Qwen3 Coder Next79.7B
  16. K2 Think V2 (this model)70B
  17. Llama 3.3 70B70B
  18. HyperNova 60B 260558.7B

Published benchmark dataset · Retrieved

Method and source
Methodology
Independent test run by Artificial Analysis on dedicated hardware.
Architecture scale
Larger models generally perform better but cost more to serve. Active parameter count matters for inference efficiency.

Output Tokens per Intelligence Index Task

Weighted average number of output tokens used to run one task in the Artificial Analysis Intelligence Index · Evaluation results measured independently by Artificial Analysis

20 models in the published order. K2 Think V2 is not marked in this published set of 20 models.
  1. Llama 4 Scout1,249.2
  2. Llama 3.3 70B1,685.9
  3. Devstral 25,946.4
  4. Muse Spark2,857.2
  5. Ling 2.6 Flash13,070.1
  6. Grok 4.3 (high)4,032
  7. Qwen3 Coder Next15,355.4
  8. Mistral Small 46,364.2
  9. GPT-5.5 (xhigh)3,572.5
  10. Qwen3.5 122B A10B6,491.8
  11. Qwen3 Next 80B A3B1,857.4
  12. Nemotron 3 Ultra7,796.7
  13. MiniMax-M311,623.4
  14. Mistral Medium 3.57,048.7
  15. NVIDIA Nemotron 3 Super7,442.1
  16. Gemini 3.5 Flash9,819.4
  17. HyperNova 60B 26055,120
  18. Claude Fable 5 (with fallback)7,695.8
  19. gpt-oss-120b (high)7,571.4
  20. DeepSeek V4 Pro (max)7,243.2

Published benchmark dataset · Retrieved

Method and source
Methodology
Independent test run by Artificial Analysis on dedicated hardware.
Benchmark interpretation
Higher values indicate stronger overall capability. Bars show how models compare on composite intelligence metrics.

Specifications

Architecture and limits

K2 Think V2 technical specifications
SpecificationValueSource
Model typeOpen weightsfaq
ReasoningYesfaq
Input modalitiesK2 Think V2 supports text input.faq
Output modalitiesK2 Think V2 supports text output.faq
Context window260k tokensfaq
Open weights / sourceYes, K2 Think V2 is open weights. The model weights are publicly available and can be downloaded for self-hosting.faq
ParametersK2 Think V2 has 70 billion parameters.faq
LicenseK2 Think V2 is released under the Apache 2.0 license. This license allows commercial use.faq

Method

Methodology & Provenance

This page is rendered from the normalized profile and page JSON for K2 Think V2.

Benchmark values are preserved as normalized; only layout, disclosure ordering, and typography are adjusted for readability.

Questions

Model FAQ and technical disclosures

K2 Think V2 was released on December 15, 2025.

K2 Think V2 was created by MBZUAI Institute of Foundation Models.

K2 Think V2 scores 17 on the Artificial Analysis Intelligence Index, placing it well above average among other open weight models of similar size (median: 9).

When evaluated on the Intelligence Index, K2 Think V2 generated 110M output tokens, which is at the higher end compared to other open weight models of similar size (median: 59M).

Yes, K2 Think V2 is a reasoning model. It uses extended thinking or chain-of-thought reasoning to work through complex problems before providing an answer.

K2 Think V2 supports text input.

K2 Think V2 supports text output.

No, K2 Think V2 does not support image input. It can only process text.

No, K2 Think V2 is not multimodal. It only supports text input.

K2 Think V2 has a context window of 260k tokens. This determines how much text and conversation history the model can process in a single request.

Yes, K2 Think V2 is open weights. The model weights are publicly available and can be downloaded for self-hosting.

K2 Think V2 has 70 billion parameters.

K2 Think V2 is released under the Apache 2.0 license. This license allows commercial use.

K2 Think V2 achieves a score of 17 on the Artificial Analysis Intelligence Index. This composite benchmark evaluates models across reasoning, knowledge, mathematics, and coding.

K2 Think V2 is an open weights model that can be self-hosted.

K2 Think V2 is an open weights model that can be downloaded and self-hosted.

Models