VerdictPal
Join

Models / Other

K2 Horizon 375B A23B

K2 Horizon · Released 2026-09-03

IFM (MBZUAI) 3 Sep 2026 flagship: 375B-A23B sparse MoE, 512K native context, Apache 2.0 weights. Official HF is IFM/K2-Horizon-375B-A23B. AA Intelligence Index v4.3 is 33.9. The 10 Sep AA snapshot lists API price as $0 / $0 — do not invent a hosted dollar rate.

Open research fleet flagship for reproducing training and running a mid-size MoE locally or via IFM partners. Prefer the AA ledger row and the IFM launch note over vendor eval tables.

#37 of 184 on AA Index · current snapshot
34 AA Index · 9 benchmark rows · 1 sourcesPosition is benchmark-specific — not a cross-family or cross-source ranking.
Composite54
Context524K
Input / 1MFree
Output / 1MFree
Knowledge cutoffunpublished
Statussolid

Family profile

Best published score in each covered benchmark family.

71/100 avg
Reasoning
87
Coding
43
Agentic
72
Long context
80

4 tested benchmark families

Editor's note
Recorded from IFM's 3 Sep 2026 launch, the official HF card (512K / 524,288 context, Apache 2.0), and the 10 Sep 2026 AA API snapshot (Intelligence Index v4.3 = 33.9, $0 / $0). IFM vendor tables stay off the ledger. Editorial depth still needs a desk pass.
Benchmark placements

Where K2 Horizon 375B A23B places on each public benchmark source that publishes a row. Rank counts every model with a latest row in the same test — not a universal quality score.

BenchmarkFamilyRankScoreSourceDate
τ³-Bench BankingAgentic#36/ 11134Artificial Analysis2026-09-10
AA-LCRLong context#38/ 18280Artificial Analysis2026-09-10
Artificial Analysis Intelligence IndexReasoning#39/ 18834Artificial Analysis2026-09-10
Terminal-BenchAgentic#44/ 18972Artificial Analysis2026-09-10
SciCodeCoding#69/ 9443Artificial Analysis2026-09-10
Humanity's Last ExamReasoning#71/ 18532Artificial Analysis2026-09-10
GPQA DiamondReasoning#75/ 19087Artificial Analysis2026-09-10
AA output speedPerformance#157/ 190Artificial Analysis2026-09-10
AA time to first tokenPerformance#157/ 190Artificial Analysis2026-09-10
Family context

The three highest-scoring models with pages in each capability family. Where K2 Horizon 375B A23B shows up, it's highlighted.

ReasoningK2 Horizon 375B A23B · 87
  1. 1
    GPT-6 Astra
    96
  2. 2
    Claude Mythos Preview
    95
  3. 3
    Gemini 3.7 Flash
    95
CodingK2 Horizon 375B A23B · 43
  1. 1
    GPT 5.5
    96
  2. 2
    Claude Opus 4.8
    96
  3. 3
    Claude Opus 4.6
    96
AgenticK2 Horizon 375B A23B · 72
  1. 1
    GPT 5.5
    100
  2. 2
    Claude Opus 4.8
    100
  3. 3
    Claude Fable 5
    100
Long contextK2 Horizon 375B A23B · 80
  1. 1
    Kimi K3
    89
  2. 2
    Claude Fable 5.1
    85
  3. 3
    GPT-5.5 (high)
    84
Newest receipts

Benchmark rows added to the public ledger for K2 Horizon 375B A23B in the last 120 days. Older rows live in the full table below.

9 recent
BenchmarkFamilyScoreSourceDays ago
AA-LCRLong context80%Artificial Analysis4
AA output speedPerformance0 t/sArtificial Analysis4
AA time to first tokenPerformance0.00sArtificial Analysis4
Artificial Analysis Intelligence IndexReasoning33.9%Artificial Analysis4
GPQA DiamondReasoning87.3%Artificial Analysis4
Humanity's Last ExamReasoning32%Artificial Analysis4
SciCodeCoding42.9%Artificial Analysis4
τ³-Bench BankingAgentic34.23%Artificial Analysis4
Terminal-BenchAgentic71.91%Artificial Analysis4
Family coverage

A high composite that hides a weak family is a trap. These bars surface the families where this model hasn't been publicly tested, and where it leads.

Reasoning87
Coding43
Agentic72
Long context80
Price vs. performance
0255075100$0.10$0.30$1$3$10$30input USD / 1M tokens · log scale →← intelligence

Hover any dot for name, score, and input price. Keyboard: tab through the top twelve, or use the ranking below.

AnthropicOpenAIMistralAlibabaDeepSeekGoogleMoonshotMetaxAIMiniMaxZhipuXiaomiOtherMeituanCohereNVIDIA

Full benchmark ledger

Every catalog benchmark for K2 Horizon 375B A23B. Scores link to the original source; gaps mean no public row exists yet.

9 sourced rows

9 of 37 catalog benchmarks have a sourced row for K2 Horizon 375B A23B.

Reasoning

Coding

Agentic

Long context

Performance

Changelog
  1. AA API snapshot: Intelligence Index v4.3 is 33.9. Hosted list still $0 / $0.
  2. IFM releases the K2 Horizon fleet under Apache 2.0, including 375B-A23B weights (512K context), data or recipes, code, and logs.
Sources