Kolibri-1 vs Inkling 2 Small

New Comparison

Scores are fetched live from the current database on every visit — nothing is cached.

Winner

Inkling 2 Small

Inkling 2 Small scores higher overall across category strength profiles (85 vs 57 for Kolibri-1); only 1 benchmarks are directly comparable.

Inkling 2 Small 85 Kolibri-1 56.9

Only 1 of 64 benchmarks are scored by every selected model — the verdict leans on category strength profiles and is indicative.

Benchmark Scores Comparison

Kolibri-1 Inkling 2 Small

Kolibri-1

78.1B params

Inkling 2 Small

594.1B params

Coverage matrix

2 models × 64 benchmarks — hover a benchmark name for its category

Full Coverage Matrix
Hides benchmarks where all selected models score identically.
Shading = score relative to the best score among the selected models: Low Mid High Missing (—) The totals row shows how many of the 64 benchmarks each model has scores for.
Benchmark Kolibri-1 Aleph Alpha Inkling 2 Small Thinking Machines
OmniScience Accuracy 14.8 —
GPQA Diamond 84.3 93
IFBench (prompt loose) 78.1 —
SealQA (12 distractors, <24k) 61 —
LiveCodeBench v6 85.9 —
BFCL-V4 61.4 —
BioMysteryBench — 70
AA-LCR 68.3 —
AIME 26 96 —
FRAMES (<24k) 71.2 —
German Public Sector 75 —
TAU3-Bench — 40.4
ScreenSpot Pro — 83.8
OmniScience Non-Hallucination 44 —
Aerospace 58.9 —
Agentic Wiki QA (DE) 69.4 —
AIME 2025 (DE) 87.5 —
Overall (EN) 75.5 —
MuirBench — 79.2
DeepSearch QA — 90.9
MuSiQue (EN) 77.3 —
SealQA (no distractors, >24k) 100 —
MMMU-Pro — 81.6
HumanEval+ 92.7 —
MMLU-ProX 75.5 —
Terminal-Bench Science 0.1 — 5.7
CharXiv (RQ) — 85.6
BrowseComp 29.4 —
BFCL v4 (web search) 62.5 —
RGB Closed-Book 51 —
Industrial Drive Technology 60 —
AA-Omniscience Index (public set) -32.8 —
SWE-bench Multilingual — 81.2
Humanity's Last Exam (DE) 15.9 —
SWE-bench Verified 66.4 —
Humanity's Last Exam 21.5 —
BFCL v4 (non-live AST) 79.1 —
BFCL v4 (live) 78.9 —
Global-MMLU-Lite — 91
BFCL v4 (memory) 62.8 —
AIME 2026 (DE) 90 —
Terminal-Bench 2.1 (Terminus-2) — 81.3
Automotive Supplier 99 —
DeepSWE — 72.1
GPQA Diamond (DE) 81.3 —
HMMT Feb 26 — 95.8
SQuAD (Utility Accuracy) 83.4 —
Terminal Bench 2.1 27.7 —
AIME 2025 96.9 —
BFCL v4 (multi-turn) 47.5 —
RGB Fact-Check (Error Correction) 34 —
SQuAD (M/A Grounding Score) 23.4 —
LongBenchPro 64.5 —
FRAMES (>24k) 73 —
Semiconductors 80.4 —
Honeypot 80.8 —
SimpleVQA — 68.2
TauBench V3 Banking 38.1 —
MCP-Atlas — 85
RGB Negative (Abstention) 85.6 —
MMLU Pro COT 80 —
Overall (DE) 70.8 —
SealQA (12 distractors, >24k) 65.1 —
SealQA (no distractors, <24k) 80.7 —
Covered 49/64 16/64

Scorecard

Overall 0–100 score — category strengths (50%), shared-benchmark wins (30%), coverage (20%)

Model Overall score Shared benchmark wins Coverage Strongest at Weakest at
Inkling 2 Small Winner 85 1 of 1 16 of 64 (25.0%) STEM & Reasoning STEM & Reasoning
Kolibri-1 56.9 0 of 1 49 of 64 (77.0%) STEM & Reasoning STEM & Reasoning

Category strengths

Where each model stands, per benchmark category — score relative to the best model, adjusted for benchmark coverage. This also ranks models when they share few benchmarks.

Category Inkling 2 Small Kolibri-1
STEM & Reasoning 13 benchmarks 100 83.1

Fully comparable benchmarks

Scored by all 2 selected models (1)

Benchmark Category Kolibri-1 Inkling 2 Small
GPQA Diamond stem_reasoning 84.3 93

Partial coverage benchmarks

Scored by at least one but not all selected models (63) — missing scores show as —

Benchmark Category Kolibri-1 Inkling 2 Small
OmniScience Accuracy 1/2 models knowledge 14.8 —
IFBench (prompt loose) 1/2 models instruction_following 78.1 —
SealQA (12 distractors, <24k) 1/2 models agentic 61 —
LiveCodeBench v6 1/2 models stem_reasoning 85.9 —
BFCL-V4 1/2 models general_agent 61.4 —
BioMysteryBench 1/2 models stem_reasoning — 70
AA-LCR 1/2 models long_context 68.3 —
AIME 26 1/2 models stem_reasoning 96 —
FRAMES (<24k) 1/2 models agentic 71.2 —
German Public Sector 1/2 models agentic 75 —
TAU3-Bench 1/2 models general_agent — 40.4
ScreenSpot Pro 1/2 models general_agent — 83.8
OmniScience Non-Hallucination 1/2 models knowledge 44 —
Aerospace 1/2 models agentic 58.9 —
Agentic Wiki QA (DE) 1/2 models agentic 69.4 —
AIME 2025 (DE) 1/2 models stem_reasoning 87.5 —
Overall (EN) 1/2 models stem_reasoning 75.5 —
MuirBench 1/2 models video_understanding — 79.2
DeepSearch QA 1/2 models general_agent — 90.9
MuSiQue (EN) 1/2 models agentic 77.3 —
SealQA (no distractors, >24k) 1/2 models agentic 100 —
MMMU-Pro 1/2 models vision_language — 81.6
HumanEval+ 1/2 models code 92.7 —
MMLU-ProX 1/2 models multilingual 75.5 —
Terminal-Bench Science 0.1 1/2 models coding_agent — 5.7
CharXiv (RQ) 1/2 models document_understanding — 85.6
BrowseComp 1/2 models general_agent 29.4 —
BFCL v4 (web search) 1/2 models agentic 62.5 —
RGB Closed-Book 1/2 models grounding_hallucination 51 —
Industrial Drive Technology 1/2 models agentic 60 —
AA-Omniscience Index (public set) 1/2 models knowledge -32.8 —
SWE-bench Multilingual 1/2 models coding_agent — 81.2
Humanity's Last Exam (DE) 1/2 models stem_reasoning 15.9 —
SWE-bench Verified 1/2 models coding_agent 66.4 —
Humanity's Last Exam 1/2 models stem_reasoning 21.5 —
BFCL v4 (non-live AST) 1/2 models agentic 79.1 —
BFCL v4 (live) 1/2 models agentic 78.9 —
Global-MMLU-Lite 1/2 models multilingual — 91
BFCL v4 (memory) 1/2 models agentic 62.8 —
AIME 2026 (DE) 1/2 models stem_reasoning 90 —
Terminal-Bench 2.1 (Terminus-2) 1/2 models coding_agent — 81.3
Automotive Supplier 1/2 models agentic 99 —
DeepSWE 1/2 models coding_agent — 72.1
GPQA Diamond (DE) 1/2 models stem_reasoning 81.3 —
HMMT Feb 26 1/2 models stem_reasoning — 95.8
SQuAD (Utility Accuracy) 1/2 models grounding_hallucination 83.4 —
Terminal Bench 2.1 1/2 models coding_agent 27.7 —
AIME 2025 1/2 models stem_reasoning 96.9 —
BFCL v4 (multi-turn) 1/2 models agentic 47.5 —
RGB Fact-Check (Error Correction) 1/2 models grounding_hallucination 34 —
SQuAD (M/A Grounding Score) 1/2 models grounding_hallucination 23.4 —
LongBenchPro 1/2 models long_context 64.5 —
FRAMES (>24k) 1/2 models agentic 73 —
Semiconductors 1/2 models agentic 80.4 —
Honeypot 1/2 models agentic 80.8 —
SimpleVQA 1/2 models vision_language — 68.2
TauBench V3 Banking 1/2 models general_agent 38.1 —
MCP-Atlas 1/2 models general_agent — 85
RGB Negative (Abstention) 1/2 models grounding_hallucination 85.6 —
MMLU Pro COT 1/2 models knowledge 80 —
Overall (DE) 1/2 models stem_reasoning 70.8 —
SealQA (12 distractors, >24k) 1/2 models agentic 65.1 —
SealQA (no distractors, <24k) 1/2 models agentic 80.7 —

Comparison FAQ

Which model is better, Inkling 2 Small or Kolibri-1?

Inkling 2 Small scores higher overall across category strength profiles (85 vs 57 for Kolibri-1); only 1 benchmarks are directly comparable.

Which model is better at STEM & Reasoning, Inkling 2 Small or Kolibri-1?

Inkling 2 Small leads the STEM & Reasoning category (strength score 100/100 across 1 comparable benchmark).

How reliable is this comparison verdict?

Only 1 of 64 benchmarks are scored by every selected model — the verdict leans on category strength profiles and is indicative.

What now?

You've seen 2 models side by side — keep the loop going.

Share this comparison

The URL is the comparison — copy or share it anywhere. No tracking, no cookies.

Not the right matchup?

Swap one model out — the pre-filled form keeps the rest of the selection.

Swap Kolibri-1 × Swap Inkling 2 Small ×