MBZUAIK2 Think V2
K2 Think V2 is a MBZUAI open-weights reasoning model, released 15 Dec 2025. Its best configuration (thinking reasoning effort) ranks #442 of 785 on the BenchLeader Index at 48.0 ±6.6, in the lower half. The ± is the point: 395 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in instruction following (63) and lowest in agents & tools (40). It has been measured at 4 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 15 Dec 2025.stale: no new result in six months
- Blended price
- –
- Output speed
- –
- First answer
- –
- Context
- 262k
- Full answer
- –
- Cost per run
- –
- Released
- 15 Dec 2025
- Overall index48
- Reasoning41
- Agents & tools40
- Knowledge43
- Instruction following63
- Long context53
- Composite42
Reasoning-effort configurations
The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| low | 42.6 | #588 | – | – | – | Agents & tools 38 · Composite 38 · Instruction following 44 · Knowledge 40 · Long context 34 · Reasoning 41 |
| medium | 45.0 | #525 | – | – | – | Agents & tools 41 · Composite 40 · Instruction following 56 · Knowledge 39 · Long context 38 · Reasoning 41 |
| high | 45.8 | #507 | – | – | – | Agents & tools 42 · Composite 41 · Instruction following 60 · Knowledge 36 · Long context 42 · Reasoning 41 |
| thinkingbest | 48.0 | #442 | – | – | – | Agents & tools 40 · Composite 42 · Instruction following 63 · Knowledge 43 · Long context 53 · Reasoning 41 |
How its index has moved
9 Sept 2026 to 23 Oct 2026- thinking
- high
- medium
- low
The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.
Benchmark results
See all LLM benchmarks. One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.
Metadata
| Benchmark | low | medium | high | thinking | Source | Trend |
|---|---|---|---|---|---|---|
| AA Openness Indexnot in index | 88.9#1 | 88.9#1 | 88.9#1 | 88.9#1 | Artificial Analysis |
Reasoning
| Benchmark | low | medium | high | thinking | Source | Trend |
|---|---|---|---|---|---|---|
| GPQA Diamond (AA)not in index | 54.1%#408 | 59.8%#378 | 68.1%#316 | 71.3%#287 | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | 3.6%#583 | 4.4%#508 | 10.5%#316 | 10.1%#324 | Artificial Analysis | |
| CritPt | 0.0%#315 | 0.0%#315 | 0.0%#315 | 0.0%#315 | Artificial Analysis |
Coding
| Benchmark | thinking | Source | Trend |
|---|---|---|---|
| Terminal-Bench 2.1 (AA)not in index | 15.0%#183 | Artificial Analysis |
Agents & tools
| Benchmark | low | medium | high | thinking | Source | Trend |
|---|---|---|---|---|---|---|
| Terminal-Bench Hard | 4.5%#284 | 8.3%#235 | 9.8%#228 | 6.8%#249 | Artificial Analysis | |
| τ²-Bench Telecom (AA)not in index | 20.8%#329 | 24.9%#296 | 27.8%#272 | 25.4%#291 | Artificial Analysis | |
| GDPval-AA v2.1not in index | – | – | – | 0.0%#124 | Artificial Analysis | |
| AA-Briefcase v1.1not in index | – | – | – | 7#66 | Artificial Analysis |
Knowledge
| Benchmark | low | medium | high | thinking | Source | Trend |
|---|---|---|---|---|---|---|
| AA-Omniscience | -47.3#365 | -49#382 | -55.6#437 | -40.3#323 | Artificial Analysis | |
| AA-Omniscience: accuracynot in index | 16.3%#393 | 17.9%#350 | 18.5%#333 | 18.0%#346 | Artificial Analysis | |
| AA-Omniscience: non-hallucinationnot in index | 24.0%#251 | 18.4%#298 | 9.0%#441 | 28.8%#226 | Artificial Analysis | |
| Global-MMLU-Lite (AA)not in index | 71.4%#31 | 76.7%#29 | 78.6%#28 | – | Artificial Analysis |
Instruction following
| Benchmark | low | medium | high | thinking | Source | Trend |
|---|---|---|---|---|---|---|
| IFBench | 41.0%#255 | 55.1%#148 | 60.1%#127 | 62.8%#117 | Artificial Analysis |
Long context
| Benchmark | low | medium | high | thinking | Source | Trend |
|---|---|---|---|---|---|---|
| AA-LCR | 20.3%#432 | 28.0%#400 | 35.0%#375 | 57.0%#293 | Artificial Analysis |
Composite
| Benchmark | low | medium | high | thinking | Source | Trend |
|---|---|---|---|---|---|---|
| AA Intelligence Index v4.3.2 | 7.3#482 | 9.0#413 | 9.9#388 | 11.5#350 | Artificial Analysis |
More from MBZUAI
See also
Questions people ask
- What are K2 Think V2's benchmark scores?
- As of 25 Oct 2026, K2 Think V2 (thinking effort) scores 48.0 ±6.6 on the BenchLeader Index, strongest in instruction following (63) and weakest in agents & tools (40). That rests on 6 independent benchmark results, and every score on this page links to its publisher.
- How does K2 Think V2 rank?
- K2 Think V2 (thinking effort) is #442 of 785 ranked configurations on 25 Oct 2026. 395 other configurations score within its margin, so the exact position matters less than the band. The leader is Claude Opus 5.5 (max effort) at 72.6.
- When was K2 Think V2 released?
- MBZUAI released K2 Think V2 on 15 Dec 2025. BenchLeader has results for it at 4 reasoning-effort settings.
Data as of 25 Oct 2026. Compare these configurations.
Cite as: BenchLeader, “K2 Think V2: benchmarks, pricing, speed and rank”, data as of 25 Oct 2026, https://www.benchleader.com/models/k2-v2.