SWE-bench-Live: Tiel solves 12 of 25, level with Opus 4.6 and ahead of KAT-Coder, Nail, stock Qwen3.6-35B-A3B and Ornith
Straight to the point
Tiel is the fast coder of the arsenal. At 4-bit quantization and 22 GB it fixes real codebase issues at the rate (and speed, with
the right GPU) of Opus 4.6 medium, while holding the best multi-turn conversation of any local model we
have measured. It is also cheerfully bad at trivia.
SWE-bench-Live: Qwen3.8-27B solves 16 of 25 at 50.2 minutes per attempt, Dirk 15 at 20.1, TielCoder 12 at 8.6, stock Qwen3.6-35B-A3B 8 at 5.5
Multi-turn conversation
Claw-Eval multi-turn: Tiel 67.2 overall against Ornith 65.3 and Nail 60.5, over 114 scored conversations each
Reasoning and knowledge
MMLU-Pro: Tiel 73.7 against Nail 84.0, both at 4-bit, with the control arm isolating the template
Where it stands. On 25 SWE-bench-Live problems Tiel fixes 12 — the same as Opus 4.6 (medium),
four more than Ornith-1.5 itself, three more than Nail, and four more than Sonnet 5 (medium). Among
models of its own class it is first; the ones ahead are dense 27Bs and Opus 5. Its time per attempt
is also steadier than Nail's: an 8.6 minute median against 7.2, but a 12.3 minute mean against 15.7,
because it lacks Nail's tail of expensive attempts.
How it talks. On Claw-Eval's multi-turn tasks Tiel scores 67.2 against Nail's 60.5 and its own
base's 65.3, over 114 scored conversations each. It earns that by answering better rather than by
asking more: against the base it is 3.8 points up on answer quality and 5.1 down on clarifying
questions. The score weights answers four to one, so the trade pays — but if you want a model that
interrogates a vague request before acting, the base does that better.
What it costs. 73.7 on MMLU-Pro against Nail's 84.0, both at 4-bit. Most of that is inherited
rather than built: Ornith-1.5 scores 78.0 where stock Qwen3.6-35B-A3B scores 85.3. Our quantization
is not the cause — the same quant carrying Ornith's own template scores exactly what Ornith scores.
The remaining 4.3 points are the Sharp template buying shorter answers, which is the trade this
build exists to make.
Which one. Agentic coding, or long conversations that have to stay useful → Tiel. Exam-style
knowledge and hard reasoning → Nail,
which is 10.3 points better on MMLU-Pro and 6.7 worse in conversation.
The most fixes per problem regardless of weight → Dirk,
the dense 27B that solves 15 of the same 25 — one behind stock Qwen3.8-27B, at 2.5x its speed.
Run it
The MoE architecture makes Tiel fast, even on smaller GPUs with partial GPU offloading, and makes the context KV small in RAM (<5 GB RAM for 262k context at 16-bit KV precision) compared to 27B dense. We do not recommend going below UD-Q4 simply to fit the whole model in GPU VRAM: when you can fit the model and context KV across your RAM+VRAM, pick a Q4 quant or larger that you can run with a sizable context (131k-262k) in at least q8_0 KV, for agentic coding.
The "fits" column below is therefore about your combined available RAM+VRAM, after OS and other processes take their share.
file
size
fits
notes
Tiel-Coder-35B-A3B-UD-Q2_K_XL.gguf
12.3 GB
16 GB
THE LAST RESORT; 2-bit gives up real ability, and struggles with agentic coding. Use anything larger, wherever it fits
Tiel-Coder-35B-A3B-UD-IQ3_XXS.gguf
13.2 GB
16 GB
the 16 GB pick — significantly better than Q2_K_XL for under a gigabyte more
Tiel-Coder-35B-A3B-UD-Q3_K_XL.gguf
16.8 GB
24 GB
3-bit with plenty of context room; prefer IQ4_XS below unless you need the extra ~1 GB
Tiel-Coder-35B-A3B-UD-IQ4_XS.gguf
17.7 GB
24 GB
4-bit quality with the most context headroom of any 4-bit tier
Tiel-Coder-35B-A3B-UD-Q4_K_S.gguf
20.9 GB
24 GB
tight 4-bit; useful when Q4_K_XL leaves too little room
Tiel-Coder-35B-A3B-UD-Q4_K_XL.gguf
22.4 GB
24-32 GB
start here — the benchmarked tier; snug on 24 GB, comfortable on 32
Tiel-Coder-35B-A3B-UD-Q5_K_XL.gguf
26.6 GB
32 GB
the 32 GB pick
Tiel-Coder-35B-A3B-UD-Q6_K_XL.gguf
31.8 GB
48 GB
near-lossless; will not leave usable context on 32 GB
Tiel-Coder-35B-A3B-UD-Q8_K_XL.gguf
38.5 GB
48 GB
reference
To download and serve, replace the gguf names in the command below with your preferred quant:
Sampling: temperature 1.0, top_p 0.95, top_k 20. For agentic coding we ran temperature 0.6.
It can see. Tiel-Coder inherits Ornith-1.5's vision tower — point it at a screenshot of a failing
test, a stack trace, a design mock. The projector is mmproj-BF16.gguf, one 903 MB file shared by
every tier, and it is Ornith's own, passed through unmodified. We changed the chat template, not the
weights, so the projector below them is theirs and a re-export of ours would be the same file.
bash
1hf download peculiar-ragdoll/Tiel-Coder-35B-A3B-GGUF \2 Tiel-Coder-35B-A3B-UD-Q4_K_XL.gguf mmproj-BF16.gguf --local-dir Tiel
3llama-mtmd-cli -m Tiel/Tiel-Coder-35B-A3B-UD-Q4_K_XL.gguf --mmproj Tiel/mmproj-BF16.gguf \4 -ngl 99 --image screenshot.png -p "Why is this test failing?"
Vision is untouched by our quantization: the projector ships at its original BF16 precision on every
tier, so a 2-bit text model and an 8-bit one see equally well.
No multi-token-prediction head
Ornith-1.5 ships an MTP (nextn) block and the tiers here do not carry it — for a reason that has
since expired, so here is the whole story rather than the conclusion.
When these tiers were baked that block was untrained. Every matrix in it sat at a standard deviation
of 0.020 with kurtosis 3.00 and a largest value of 5 standard deviations, which is a fresh random
initialization, against kurtosis between 4 and 39 with outliers past 14 in every trained layer of the
same file. Drafts from random weights are accepted at chance, so it was 2.4% of every file doing no
work, and removing it changed no output.
Ornith fixed it on 2026-08-23, re-uploading a single shard with a trained head. We measured the
replacement rather than take it on faith: kurtosis 25.1 with a 98-sigma outlier, and 825 on the
nextn projection. That is not fresh initialization by any reading. The head is real now, and it
ships in Tiel-Coder-35B-A3B-GGUF-MTP.
The tiers in this repo are current and were not rebuilt. The re-upload touched only the MTP
tensors — lm_head is bit-identical across the fix and layer 39's experts have an identical value
multiset — so nothing else moved. Take the MTP repo if your runtime does multi-token speculative
decoding; take this one if it does not, and keep the 0.9 GB the head costs.
How the imatrix was made
Every tier is quantized against an importance matrix we generated ourselves, rather than a
borrowed one. The calibration corpus is 49 M characters drawn from
eaddario/imatrix-calibration (MIT):
about three quarters code_medium and one quarter combined_all_large, interleaved so the two
alternate throughout rather than sitting in separate halves. That mix is deliberate — Tiel is meant
for coding, so the corpus leans that way, while the combined slice keeps maths, tool-calling and
non-English text represented so those paths are not the ones that get quantized carelessly.
The matrix was measured on a Q8_0 of the original BF16 weights, over 3,000 chunks of 512 tokens
(~1.5 M tokens), which is enough for every expert to be exercised many times over — this is a
256-expert mixture that routes 8 per token, so a short corpus would leave some experts barely seen.
The shipped tiers are then quantized from the BF16 source using that matrix.
The matrix itself ships here, as Tiel-Coder-35B-A3B.imatrix.gguf (183 MiB, 510 tensors) — so a
tier we don't ship is one command away, without spending an hour and a half measuring your own:
Be clear about what that does not give you: the tiers in the table above are cut with per-tensor
Dynamic recipes layered on top of this matrix, and they carry the Sharp template. A plain
llama-quantize from the upstream BF16 reproduces neither.
