Cataloged from peculiar-ragdoll/Tiel-Coder-35B-A3B-GGUF
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.
Pick it for work. Pick something else for exams.
This is Ornith-1.5-35B-A3B re-quantized dynamically with our own imatrix and carrying the Sharp chat template inside the GGUF. Find TielCoder MTP GGUFs here.
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.
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:
hf download peculiar-ragdoll/Tiel-Coder-35B-A3B-GGUF \
Tiel-Coder-35B-A3B-UD-Q4_K_XL.gguf --local-dir Tiel
llama-server -m Tiel/Tiel-Coder-35B-A3B-UD-Q4_K_XL.gguf -ngl 99 --jinja
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.
hf download peculiar-ragdoll/Tiel-Coder-35B-A3B-GGUF \
Tiel-Coder-35B-A3B-UD-Q4_K_XL.gguf mmproj-BF16.gguf --local-dir Tiel
llama-mtmd-cli -m Tiel/Tiel-Coder-35B-A3B-UD-Q4_K_XL.gguf --mmproj Tiel/mmproj-BF16.gguf \
-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.
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.
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:
hf download peculiar-ragdoll/Tiel-Coder-35B-A3B-GGUF Tiel-Coder-35B-A3B.imatrix.gguf --local-dir .
llama-quantize --imatrix Tiel-Coder-35B-A3B.imatrix.gguf Ornith-1.5-35B-BF16.gguf out.gguf IQ4_XS
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.
eaddario's code slice is itself built from Open-Critic-GPT, opc-sft-stage2, Magicoder-Evol-Instruct-110K and McEval-Instruct.
MIT, inheriting Ornith-1.5's license.