LFM2-2.6B vs Qwen3 4B 2507 Instruct
Wins 0 of 6 areas
—
Wins 3 of 6 areas
Coding · Long documents · Following instructions
Qwen3 4B 2507 Instruct wins more areas, narrowly.
Scores updated · 24 tests both models report · How we compare
Where each one wins
Tests won in each of the six areas where both have results. Each piece is one test, so a longer bar means more evidence; grey means the two scored within a point of each other.
- CodingWriting and fixing software03Qwen3 4B 2507 Instruct3 of 3 tests
- Long documentsFinding answers in very long texts01Qwen3 4B 2507 Instruct1 of 1 test
- Following instructionsDoing exactly what it is asked01Qwen3 4B 2507 Instruct1 of 1 test
- AgentsCarrying out multi-step tasks on its own00Even0 each · 1 tie
- ReasoningHard problems that need careful thinking11Even1 each · 1 tie
- FactsGetting facts right instead of making them up11Even1 each
Agents, long documents and following instructions rest on a single test each.
What it costs
Prices per million tokens, roughly 750,000 words. The bars show the cost of a million tokens read plus a million written.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where Qwen3 4B 2507 Instruct pulls ahead
- Recent programming contest problemsLiveCodeBench v6+34.3points ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+21.1points ahead
- Code for real scientific research problemsSciCode+15.6points ahead
Where LFM2-2.6B pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+18.8points ahead
- Very hard expert questions across many subjectsHumanity's Last Exam+1points ahead
Every test, side by side
All 24 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingQwen3 4B 2507 Instruct
- LiveCodeBench v6Qwen3 4B 2507 Instruct by 34.314.448.7+34.3
- SciCodeQwen3 4B 2507 Instruct by 15.62.518.1+15.6
- Terminal-Bench HardQwen3 4B 2507 Instruct by 3.70.84.5+3.7
ReasoningEven
- GPQA DiamondQwen3 4B 2507 Instruct by 21.130.651.7+21.1
- Humanity's Last ExamLFM2-2.6B by 15.54.5+1
- CritPttie00tie
FactsEven
- AA-Omniscience · Non-hallucinationLFM2-2.6B by 18.837.118.3+18.8
- AA-Omniscience · AccuracyQwen3 4B 2507 Instruct by 5.55.410.9+5.5
Long documentsQwen3 4B 2507 Instruct
- AA-LCRQwen3 4B 2507 Instruct by 11.3011.3+11.3
Following instructionsQwen3 4B 2507 Instruct
- IFBenchQwen3 4B 2507 Instruct by 1419.533.5+14
Other results13 tests, not counted
Tests outside the eight areas. They are not counted above: several are summary scores built from other tests, or the same test under another name.
- LCB v5Qwen3 4B 2507 Instruct by 36.414.450.8+36.4
- MMLU-ProQwen3 4B 2507 Instruct by 26.32652.3+26.3
- HumanEval+Qwen3 4B 2507 Instruct by 24.457.982.3+24.4
- MATH-500 (EM)Qwen3 4B 2507 Instruct by 2263.685.6+22
- GSM8KLFM2-2.6B by 13.982.468.5+13.9
- τ²-Bench Telecom (AA run)Qwen3 4B 2507 Instruct by 13.113.526.6+13.1
- MMLUQwen3 4B 2507 Instruct by 7.964.472.3+7.9
- AA-OmniscienceLFM2-2.6B by 7.8-54.1-61.9+7.8
- MGSMQwen3 4B 2507 Instruct by 7.574.381.8+7.5
- IFEvalQwen3 4B 2507 Instruct by 679.685.6+6
- MMMLUQwen3 4B 2507 Instruct by 5.355.460.7+5.3
- AA Agentic IndexQwen3 4B 2507 Instruct by 4.44.58.9+4.4
- AA IntelligenceQwen3 4B 2507 Instruct by 1.55.26.7+1.5
Questions people ask
Which is better, LFM2-2.6B or Qwen3 4B 2507 Instruct?
Qwen3 4B 2507 Instruct wins three of the six areas where both have results: coding, long documents and following instructions. LFM2-2.6B wins none. They are level on agents, reasoning and facts.
Which is better for coding?
Qwen3 4B 2507 Instruct. It wins 3 of the 3 coding tests both models report; LFM2-2.6B wins none.
How do you compare the two?
We use the 24 benchmark tests both models have published scores on. The verdict counts the 11 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 13 are listed but not counted, because several are summary scores or repeat a test. Each score is the one shown on the model's own page.