Provider evidence · synced runs

The record behind the score.

KYBER publishes the newest valid provider strike for each model. Every contract below carries its suite, rules, filter, sample plan, and source file. Coverage determines whether it can enter the official leaderboard.

Synced models

one newest strike per model

Ranked

complete current suite

Suite fingerprint

1adbc0d3

1adbc0d3703acf03

Published contracts

1,500 trials in full coverage

Rank 01 · leaderboard eligible

rustlean

2026-10-01 · 1500/1500 current suite trials · 41.5% pass@1

score 0.432weighted 0.474valid-pass 41.5%1 sample

Inspect contract

Source: kyber_rustlean.jsonl
Engine: 0.1.0
Suite fingerprint: 1adbc0d3703acf03
Oracle rules: bcdc6c167e45926c
Elicitation: 98459811a25563ac
Response protocol: schema-object-text-v2
Requested format: auto
Actual formats: json_schema × 1,500
Filter: full
Plan: declared 1500 · sample 0 · limit 0
Sampling: 1 answers · temp 0 · jobs 1

Model report Browse trial evidence

Rank 02 · leaderboard eligible

lfm2.5-2.6b-fable5-coding-agent-heretic

2026-09-22 · 1500/1500 current suite trials · 34.3% pass@1

score 0.345weighted 0.348valid-pass 34.3%1 sample

Inspect contract

Source: kyber_lfm2_5-2_6b-fable5-coding-agent-heretic.jsonl
Engine: 0.1.0
Suite fingerprint: 1adbc0d3703acf03
Oracle rules: bcdc6c167e45926c
Elicitation: b561b68a6ad11490
Response protocol: legacy · JSON object / text
Requested format: not recorded
Actual formats: not recorded
Filter: full
Plan: declared 1500 · sample 0 · limit 0
Sampling: 1 answers · temp 0 · jobs 12

Model report Browse trial evidence

Rank 03 · leaderboard eligible

minicpm5-2b-claude-fable5-1-thinking-agentic

2026-09-22 · 1500/1500 current suite trials · 27.8% pass@1

score 0.296weighted 0.283valid-pass 27.8%1 sample

Inspect contract

Source: kyber_minicpm5-2b-claude-fable5-1-thinking-agentic.jsonl
Engine: 0.1.0
Suite fingerprint: 1adbc0d3703acf03
Oracle rules: bcdc6c167e45926c
Elicitation: b561b68a6ad11490
Response protocol: legacy · JSON object / text
Requested format: not recorded
Actual formats: not recorded
Filter: full
Plan: declared 1500 · sample 0 · limit 0
Sampling: 1 answers · temp 0 · jobs 4

Model report Browse trial evidence

Rank 04 · leaderboard eligible

minicpm5-1b-claude-opus-fable5-v2-thinking

2026-09-22 · 1500/1500 current suite trials · 18.1% pass@1

score 0.184weighted 0.169valid-pass 18.1%1 sample

Inspect contract

Source: kyber_minicpm5-1b-claude-opus-fable5-v2-thinking.jsonl
Engine: 0.1.0
Suite fingerprint: 1adbc0d3703acf03
Oracle rules: bcdc6c167e45926c
Elicitation: b561b68a6ad11490
Response protocol: legacy · JSON object / text
Requested format: not recorded
Actual formats: not recorded
Filter: full
Plan: declared 1500 · sample 0 · limit 0
Sampling: 1 answers · temp 0 · jobs 4

Model report Browse trial evidence

Publication rule

A rank is earned on the whole suite.

The leaderboard accepts a current engine and oracle fingerprint, the full unfiltered suite, no sampling or limit, and one deduplicated result for every trial. Partial runs remain inspectable here and on the trial browser; they never compete against complete runs.

See the official top 25