VerdictPal · editorial desk · 2026VerdictPal

Compare · Two models, head-to-head

Pick two models. See the diff.

Composite scores, benchmark ranks, capability grids, and pricing per million tokens. All sourced from the same data that powers the model atlas.

Cards Models

Kimi K2.5 vs MiniMax M3

Model A

Model B

MODEL A

Kimi K2.5

Moonshot · Kimi

Composite 53Intel 3513 rows4 sources

Moonshot's January 2026 multimodal open-weights model. 262K context, $0.60/$3 per 1M tokens. Agent Swarm up to 100 sub-agents. Predecessor to K2.6.

MODEL B

MiniMax M3

MiniMax · MiniMax

Composite 62Intel 4417 rows8 sources

MiniMax's May 2026 open-weights release. Terminal-Bench 2.0 66.0% — the #10 score globally and the strongest non-frontier-tier result on the public ledger. 1M context, $0.40 / $2.00 per 1M tokens.

Kimi K2.51stats won

Tied1stats

MiniMax M35stats won

Spec showdown

Capacity, price, freshness.

Kimi K2.5vsMiniMax M3

MiniMax M3 holds a narrow composite edge (62 vs 53), but the gap is small enough that other factors matter more. MiniMax M3 has the higher Vals/AA intelligence score (44 vs 35). MiniMax M3's 1M context window dwarfs Kimi K2.5's 262K — a meaningful difference for long-document or codebase work.

Composite score
Kimi K2.553MiniMax M362Winner on this row
Intelligence (Vals/AA)
Kimi K2.535MiniMax M344Winner on this row
Context window
Kimi K2.5262KMiniMax M31MWinner on this row
Max output
Kimi K2.566KMiniMax M366K
Input price / 1M (lower wins)
Kimi K2.5$0.60MiniMax M3$0.40Winner on this row
Output speed (AA default API)
Kimi K2.550 t/sMiniMax M397 t/sWinner on this row
Time to first token (AA default API)
Kimi K2.51.02sWinner on this rowMiniMax M31.26s

Composite score

Intelligence (Vals/AA)

Context window

262K

Max output

66K

Input price / 1M (lower wins)

$0.60

$0.40

Output speed (AA default API)

50 t/s

97 t/s

Time to first token (AA default API)

1.02s

1.26s

Capability grid

What each model can do.

Modalities

Kimi K2.5MiniMax M3

Text●●

Image●—

Audio——

Video●—

Code●●

Tool use●●

Reasoning●●

Access

Kimi K2.5MiniMax M3

API●●

Consumer app——

Open weights●●

Managed cloud●●

On-device——

Family radar

Top score per benchmark family.

Knowledge

Reasoning

Coding

Agentic

Long context

Benchmark ranks

Every shared benchmark, side by side.

Knowledge

MMLU#18/22 · #22/22

86.4%

84.8%

Reasoning

Artificial Analysis Intelligence Index#44/67 · #21/67

35.4%

44.4%

GPQA Diamond#32/69 · #7/69

87.9%

92.9%

Humanity's Last Exam#37/65 · #23/65

29.4%

37.1%

IFBench#35/57 · #1/57

70.2%

82.86%

LiveBench#16/21 · #18/21

61.2%

59.6%

Coding

SciCode#31/65 · #43/65

49%

45.4%

Agentic

Terminal-Bench#48/69 · #28/69

45.69%

66%

τ³-Bench Banking#48/64 · #51/64

14.23%

12.99%

Long context

AA-LCR#39/64 · #7/64

65.33%

74%

LongBench v2#15/19 · #18/19

48.2%

45.6%

Performance

AA output speed#55/69 · #26/69

50 t/s

97 t/s

AA time to first token#43/69 · #38/69

1.02s

1.26s

Dossier facts

Dates, sources, fineprint.

Kimi K2.5MiniMax M3

ProviderMoonshotMiniMax

FamilyKimiMiniMax

Release date2026-01-272026-05-31

Knowledge cutoff2025-10-012026-03-01

Context window262K1M

Max output66K66K

Input price /1M$0.60$0.40

Output price /1M$3.0$2.0

Statussolidsolid

Composite score5362

Intelligence score3544

Intel sourceaaaa

Benchmark rows1317

Source count48

Freshnessfreshfresh

Pricing checked2026-06-092026-06-01

Ecosystem

Which tools wrap each model.

Kimi K2.5

Not listed in any tool yet.

MiniMax M3

Not listed in any tool yet.

Verdict

Which model to actually pick.

Open Kimi K2.5's dossier Open MiniMax M3's dossier