Empero — Independent AI research labOpen weights01/06<br>Menu<br>EMPERO/00 — INDEXINDEPENDENT AI RESEARCH LAB · OPEN BY DEFAULT · BUILT IN GERMANY2026.06.27<br>Independent AI research lab<br>Small models,<br>trained in the open.<br>An independent AI research lab. We build small, efficient language models you can own and run yourself — and release the weights, code and datasets in the open. This release wave ships Abacus , our Rust terminal coding agent; refreshes the Qwythos GGUFs with v2 runtime fixes and MTP variants; and announces Qwythos-27B as the next larger Mythos model. Claire remains our in-house language model in training.<br>See the models →Abacus on GitHub ↗The research →

Abacus<br>Coding agent<br>released today · Rust TUI

GGUF v2<br>Qwythos local builds<br>fixed templates · MTP · vision

27B<br>Qwythos announced<br>larger Mythos tier

100%<br>Open by default<br>weights · code · datasets

EMPERO — INDEPENDENT AI RESEARCH LAB✱ABACUS · CODING AGENT · RELEASED TODAY✱QWYTHOS-9B · CLAUDE MYTHOS 5 DISTILL · 1M CONTEXT✱QWYTHOS GGUF V2 · FIXED TEMPLATES · MTP VARIANTS · VISION✱QWYTHOS-27B · ANNOUNCED✱CLAIRE · IN-HOUSE MODEL · 6B-A500M · IN TRAINING✱MICROVERSE · AUTOMATED ARCHITECTURE DISCOVERY✱OPEN WEIGHTS · OPEN CODE · OPEN DATASETS✱EMPERO — INDEPENDENT AI RESEARCH LAB✱ABACUS · CODING AGENT · RELEASED TODAY✱QWYTHOS-9B · CLAUDE MYTHOS 5 DISTILL · 1M CONTEXT✱QWYTHOS GGUF V2 · FIXED TEMPLATES · MTP VARIANTS · VISION✱QWYTHOS-27B · ANNOUNCED✱CLAIRE · IN-HOUSE MODEL · 6B-A500M · IN TRAINING✱MICROVERSE · AUTOMATED ARCHITECTURE DISCOVERY✱OPEN WEIGHTS · OPEN CODE · OPEN DATASETS✱

Now shipping<br>Abacus, Qwythos GGUF v2, and Qwythos-27B.

June 22, 2026<br>ABACUS · RELEASED TODAY<br>Abacus is the coding agent.<br>A fast, local-first terminal agent in Rust for setup, search, edits, review, sessions and scripting. Bring your own model endpoint; every mutation is approval-gated and shown first as a per-file diff.

Rust · local/open-weight friendlyOpen on GitHub ↗<br>QWYTHOS GGUF · V2<br>Redownload the Qwythos GGUFs.<br>v2 replaces the original normal files, fixes tokenizer and embedded chat/tool-template metadata for Qwen3.5 GGUF runtimes, adds -MTP- variants, and smoke-tests Q4/Q8 tool calling, 1M context and vision.

Q4_K_M through BF16 · MTP + mmprojOpen GGUF v2 ↗<br>QWYTHOS-27B · ANNOUNCED<br>The next Qwythos size is on deck.<br>Qwythos-27B is announced as the larger next member of the Mythos line. The shipped 9B remains the current open-weight flagship while 27B moves the family toward a heavier local-reasoning tier.

Announced · details to followFollow on HF ↗

01Flagship · Qwythos-9BQWEN3.5-9B · GGUF V2 · 1M CONTEXT · APACHE-2.0<br>CURRENT FLAGSHIP · GGUF V2 LIVE · QWYTHOS-27B ANNOUNCED<br>A 9B that checks<br>its own work.<br>Qwythos-9B-Claude-Mythos-5-1M is our shipped flagship — a full-parameter reasoning model distilled from Claude Mythos 5 and Claude Fable traces, with a 1M-token context , native tool use, and a +34.3 MMLU lift over its base. The v2 GGUFs are live now: fixed Qwen3.5 runtime metadata, stronger embedded chat/tool templates, normal replacement files, explicit -MTP- variants for draft speculation, and a vision projector. Qwythos-27B is announced as the next larger Mythos model.

Model card on HF ↗GGUF v2 for llama.cpp / Ollama ↗Full eval transcripts ↗

+34.3<br>MMLU vs base<br>0.232 → 0.575

1M<br>token context<br>YaRN · default

v2<br>GGUF refresh<br>templates · MTP · vision

27B<br>announced next<br>larger Mythos tier

Matched lm-evaluation-harness run (limit 100). Not every benchmark moved up — gpqa-diamond −0.05, arc-challenge roughly flat. Full table on the model card.<br>Reason → tool → verify · real transcripts, condensedfrom the published eval set<br>MATH · PYTHON TOOLHow many prime numbers are there below 100,000? Verify with python.+PROMPTHow many prime numbers are there below 100,000? Verify with python.<br>TOOL CALLS›python_executor → wrote and ran a primality test → stdout: 9592

· REASONING · 2 rounds · 10.5sThis is a counting problem over a fixed range — the kind of thing I should verify rather than recall. I'll write a primality test, count every prime from 2 through 99,999, and run it in the python_executor rather than trusting a memorised figure.<br>FINAL ANSWERThere are 9,592 prime numbers below 100,000.

I verified this by computing all primes from 2 through 99,999 using a primality test and counting them.

→ Reaches for a tool to verify rather than recalling a number. Writes the primality test, runs it, reports the checked result.

CYBERSECURITY · WEB SEARCHWhat is the hashcat hash-mode (-m) number for cracking a Kerberos TGS-REP ticket extracted via Kerbe…+<br>CLINICAL · SAFETY-CRITICALIs physostigmine indicated for organophosphate (nerve agent) poisoning? Or is it used for a differen…+

02Published models6 ON HUGGING FACE · OPEN WEIGHTS<br>CURRENT FLAGSHIP · 9B · 1M ctxQwythos-9B-Claude-Mythos-5-1MCLAUDE FABLE 5 DISTILL · 9BQwable-9B-Claude-Fable-5QWEN DISTILL · 9BQwen3.5-9B-Claude-Opus-4.6-DistillOPENNEMO · PORT · 9BopenNemo-9BOPENNEMO · PORT · 30B (3B active)openNemo-Cascade-2-30B-A3BQWEN...