Runs on ≤96 GB · Hugging Face ↗
Not independently scored by Artificial Analysis
Released 2026-09-01 by the Institute of Foundation Models (MBZUAI), Apache-2.0. The largest dense member of the K2 Horizon family: 34.8B parameters, 64 layers, 512K declared context.
BigCodeBench-Hard pass@1 32% (47/148), via official protocol · llama.cpp Q8_0 · one request at a time · a shared cluster GPU · mean of 3 runs, range 0.0 pp. The brutal counterpart to HumanEval — where HumanEval saturates near the top, BCB-Hard spreads the field, so this is the number that actually separates coding ability. 1 of 148 answers hit the token limit (effective ceiling 99%)
Our own runs on this model. Each number carries how many problems it is over and how many times we ran it — these are not comparable with each other, and none of them is on the same scale as the third-party indices on the leaderboard.
| test | score | over | conditions |
|---|---|---|---|
| BigCodeBench-Hard | 31.8% | n=148 · 3 run(s) | official protocol · llama.cpp Q8_0 · one request at a time · a shared cluster GPU · mean of 3 runs, range 0.0 |
| RULER (long context) | 94.9 ± 4.7 | N=5/task · at 32K | 2 quantisations measured |
| quant | GiB | code pass@1 | 24 GB | 32 GB | 48 GB | 96 GB | 141 GB |
|---|---|---|---|---|---|---|---|
| Q8_0 | 34.4 | 31.8% | — | — | 152K / 80K | 524K / 432K | 524K / 524K |
| Q4_K_M | 19.6 | 34.0% | 36K / 19K | 147K / 77K | 368K / 195K | 524K / 524K | 524K / 524K |
We have not been able to tie these files back to a published repo, so take the quant names as the labels we ran them under. Every point measured on the same class of card, one pass, one request at a time. One problem is 0.68 points, so anything under ~3.4 points apart is a tie — that is the spread we measure between GPU classes, and our rows are not all on the same card. Each card column is the context left for the KV cache after the weights, KV-Q4 / KV-Q8. A dash means the file does not fit that card, or fits with no room left to work. Context is usually the real constraint, not quality. Computed for one slot (--parallel 1), which is what you get serving yourself. llama-server opens four by default, and on a hybrid that costs real context — the table below each curve shows how much. The card figures assume a dedicated card. Measured on an idle one: the driver keeps 623 MiB of 143,771, so 99.6% is usable. Compute buffers are measured too — a fixed ~130 MiB plus 1 MiB per 1K of context, the same slope on every architecture we checked. If your card also runs your desktop you get noticeably less, and we have not measured that case yet.
| quant | GiB | 4K | 8K | 32K |
|---|---|---|---|---|
| Q8_0 | 34.4 | 98.0 ± 3.2 · N=5 | 96.4 ± 4.2 · N=5 | 94.9 ± 4.7 · N=5 |
| Q4_K_M | 19.6 | 98.0 ± 3.2 · N=5 | 100.0 ± 0.0 · N=5 | 94.9 ± 4.7 · N=5 |
All 13 RULER tasks, one request at a time, KV cache f16, bench commit c3f5e3b4, seed 42. The score is the mean of the 13; the margin is two standard deviations of the sampling, so it grows as the sample shrinks and as the score drops. Everything here clears the paper’s effective-length bar of 85.6 at every length shown.
This is not a coding score and must not be read as one. RULER asks whether the model finds and follows a fact buried in a long text. On this model Q8_0 scores 94.9 at 32K against 94.9 for Q4_K_M, a difference of 0.0 points — while on BigCodeBench-Hard the same two files score 31.8% and 34.0%. A page that read the long-context number as a general score would be telling the truth and misleading you.
Real GGUF file sizes = weight VRAM. Add the KV cache for your context (≈9 GB at 32K, fp16). Read from the GGUF header, but this architecture has never been checked against llama.cpp — treat it as an estimate.
| Quantization | Size |
|---|---|
| Q4_K_M | 21.1 GB |
| Q5_0 | 24.1 GB |
| Q5_K_M | 24.7 GB |
| Q6_K | 28.5 GB |
| Q8_0 | 37.0 GB |
| BF16 | 69.6 GB |
Upstream llama.cpp did not load it when we measured: we used a build of IFM's own branch (model/K2Horizon, commit 42adf01). IFM only publishes BF16 (69.6 GB), so the Q8_0 and Q4_K_M points in our curve are quantised by us from that file with llama-quantize, using the same build, with no importance matrix. Artificial Analysis does not score it.