Request access to the BottleCap AI model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Request access below. Add your company if you're evaluating this for work — we have enterprise versions that go further, and we'll make sure you hear about them first. We'll also send you new ThinkingCap releases and early access before they're public.

Tell us how you plan to use the model and we can help you get the most out of it — there's a short form for that too.

ThinkingCap-Qwen3.8-27B-FP8

FP8 (e4m3) block-wise weights, 128x128 blocks, dynamic activation scales; loads in vLLM and SGLang, native on Hopper and Blackwell.

Built from bottlecapai/ThinkingCap-Qwen3.8-27B (bf16). Vision tower, MTP head, lm_head and the GDN in_proj_a / in_proj_b projections stay bf16 (quantization_config.ignore in config.json). Serve with vLLM 0.29 (--trust-remote-code not needed).

Expected performance

Paired comparison with the bf16 source on the full quantization plan: both builds answer the same questions with the same seeds — RealWorldQA 765 questions × 2 seeds (images), GPQA-Diamond 198 × 4, MMLU-Pro 1,500 × 1 (a fixed slice), IFBench 300 × 2, AA-LCR 100 × 1 (long-document prompts, graded by Gemma-4-26B-A4B-it with thinking off). Thinking at the chat template's default reasoning effort (xhigh), sampled decoding (temperature 1.0, top_p 0.95, top_k 20, min_p 0.0), 65,536-token generation cap. Both builds served by vLLM 0.29.0 on an H200.

benchmark (questions × seeds)	accuracy %, bf16 → FP8	Δ accuracy, pp [95% CI]	tokens mean / median / p95, bf16 → FP8	Δ mean tokens [95% CI]	Δ median tokens
RealWorldQA (765 × 2)	83.1 → 83.6	+0.5 [−1.0, +2.0]	488 / 112 / 1,912 → 550 / 113 / 2,722	+12.8% [−4.3, +32.8]	+0.9%
GPQA-Diamond (198 × 4)	88.0 → 87.8	−0.3 [−2.2, +1.7]	7,115 / 1,031 / 37,459 → 7,084 / 1,220 / 38,326	−0.4% [−8.3, +7.7]	+18.4%
MMLU-Pro (1,500 × 1)	84.1 → 84.1	0.0 [−1.3, +1.3]	1,436 / 166 / 7,255 → 1,323 / 176 / 6,977	−7.9% [−19.2, +6.0]	+6.0%
IFBench (300 × 2)	79.7 → 79.3	−0.3 [−3.6, +2.9]	4,531 / 1,822 / 20,630 → 4,283 / 1,644 / 18,008	−5.5% [−13.1, +3.0]	−9.8%
AA-LCR (100 × 1)	81.0 → 82.0	+1.0 [−5.5, +7.5]	1,718 / 844 / 4,937 → 1,506 / 855 / 4,593	−12.4% [−30.8, +10.0]	+1.2%

Δ accuracy is FP8 minus bf16 on the same answers; its interval treats the question as the unit (seeds averaged per question first). Tokens are completion tokens (reasoning plus answer).

Throughput — one RTX PRO 6000 Blackwell, vLLM 0.29.0, synthetic prompts of 1,024 tokens with 512 generated (last column: 32,768-token prompts, 128 generated), end-of-sequence ignored, prefix caching off, --max-num-seqs 64. Aggregate output tokens/s, median time to first token in ms in parentheses.

build	1 request	16 concurrent	64 concurrent	4 concurrent, 32k-token prompts
bf16	26.2 (158)	341 (1,867)	844 (3,627)	18.6 (9,678)
FP8	44.7 (107)	540 (1,227)	1,227 (2,369)	27.0 (7,668)

Decode speed and MTP self-speculative decoding (MMLU-Pro) — 32 questions × 1 seed, vLLM 0.26 on one H200, 16 concurrent requests

config	median tokens	tok/s	s / task	MTP speedup	accept_len (max 4)
Qwen3.8-27B base · standard	484	51.3	8.9	1.00×	—
Qwen3.8-27B base · MTP	502	91.0	4.2	1.77×	2.59
ThinkingCap-Qwen3.8-27B bf16 · standard	232	50.5	3.7	1.00×	—
ThinkingCap-Qwen3.8-27B bf16 · MTP	216	85.7	2.0	1.70×	2.60
FP8 · standard	216	62.9	2.8	1.00×	—
FP8 · MTP	252	112.1	2.3	1.78×	2.55

Where to find us

Need even more efficiency? The open release is production-ready. Our enterprise versions go further — fewer thinking tokens still, tuned to your workload, at matched accuracy on your own tasks. Built for AI labs, inference providers and enterprises running models at scale. Deployed on your infrastructure, or in the cloud and region you choose. Talk to our team

License

ThinkingCap: PolyForm Small Business 1.0.0 + BottleCap personal-use grant (see LICENSE).

Upstream Qwen materials: Apache-2.0 (see NOTICE).

Commercial license: contact BottleCap AI.

Citation

If you use this model, please cite:

@misc{ThinkingCap-Qwen3.8-27B,
  title     = {bottlecapai/ThinkingCap-Qwen3.8-27B},
  author    = {Osusky, Adam and Lindauer, Jan and Jirkovsky, Adam and Mihal, Filip and Platek, Ondrej and Herel, David and Ihnatchenko, Luka and Bartek, Vojtech and Jirak, Jiri and Kubista, Daniel and Krus, Frantisek and Mikolov, Tomas},
  year      = {2026},
}