Model result, rank 26 of 27

Qwythos‑9B Claude Mythos Q8_0

Local model

Local GGUF · llama.cpp Vulkan · Q8_0 · 256K allocation verified. Public result card with the model’s overall score, lane measurements, runtime and cost telemetry, and the ranking formula.

Compare all models Download public JSON

Figure 1 Lane profile against the cohort Each bar is one measured lane on the shared 0 to 100 scale. The thin mark is the cohort median; the bright mark is the best score any ranked model reached on that lane.

Full / Agenticlane rank #25 of 27

68.15

SWE MVPlane rank #26 of 27

46.51

Hard Intelligencelane rank #26 of 27

43.91

Qwythos‑9B Claude Mythos Q8_0
Cohort median
Cohort best

Overall score52.86

Rank #26

Full / Agentic68.15

Full rank #25

SWE MVP46.51

SWE rank #26

Hard Intelligence43.91

Hard rank #26

Measured cost$0

100.0% reliability

Overall

All-around publication view

Score52.86

Formulamean(Full, SWE, Hard Intelligence)

BasisLocal GGUF · llama.cpp Vulkan · Q8_0 · 256K allocation verified

The overall score averages the measured major lanes while keeping each source measurement visible.

Lane 01

Full / Agentic benchmark

Final68.15

Capability83.82

Agentic76.23

Pass rate80.0%

Prompts43

This lane captures instruction following, structured behavior, tool discipline, and general agentic reliability.

Lane 02

Software engineering MVP

SWE score46.51

Focused final46.51

Capability25.00

Daily driver53.25

Prompts24

This lane is closer to implementation usefulness: source handling, architecture cleanliness, and deliverable quality.

Lane 03

Hard Intelligence diagnostic

Hard score43.91

Active inquiry81.42

Online adaptation6.88

Self-repair78.33

Authority integrity9.00

Hard Intelligence measures active inquiry, online adaptation, evidence-driven self-repair, and authority/salience integrity.

Telemetry

Runtime economics

Total cost$0

Cost / scored item$0

Seconds / timed item21.13s

Runtime coverage100.0%

Recorded tokens / item6.0k

Token coverage100.0%

Cost, time, and token basis are normalized telemetry. They explain tradeoffs; they do not overwrite the capability score yet.

Interpretation

Why the result lands here.

The model is stronger in the Full/Agentic lane than in the SWE lane; the overall score is therefore shown with both component lanes visible. Hard Intelligence score is 43.91 and contributes to the overall score alongside Full/Agentic and SWE. Local model row: benchmarked on local hardware with no API metering. 256K allocation and runtime fit were verified for this Q8_0 local Vulkan entrant; filled-context retrieval at 256K remains separate/unproven. Q8 shows stronger full-suite quality than Q6, but SWE and Hard Intelligence diagnostic lanes remain weak.