The two are evenly matched.Celeris-1 is better at reasoning and coding; G9v3-3B at agents and long documents.
Scores updated · 13 tests both models report · How we compare
Where each one wins
Tests won in each of the five 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.
- ReasoningHard problems that need careful thinking20Celeris-12 of 3 tests · 1 tie
- CodingWriting and fixing software20Celeris-12 of 2 tests
- AgentsCarrying out multi-step tasks on its own02G9v3-3B2 of 2 tests
- Long documentsFinding answers in very long texts01G9v3-3B1 of 1 test
- FactsGetting facts right instead of making them up11Even1 each
Long documents rests on a single test.
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 three tests each model wins by the widest margin. Scores are out of 100.
Where Celeris-1 pulls ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+19.3points ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+5.2points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+4.5points ahead
Where G9v3-3B pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+81.1points ahead
- Reasons across sets of long documentsAA-LCR+3.3points ahead
- Customer-service tasks in a simulated bankτ-Bench V3 · Banking+1.9points ahead
Every test, side by side
All 13 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingCeleris-1
- Terminal-Bench 2.1Celeris-1 by 5.211.26+5.2
- SciCodeCeleris-1 by 3.921.617.7+3.9
AgentsG9v3-3B
- GDPValG9v3-3B by 15.4015.4+15.4
- τ-Bench V3 · BankingG9v3-3B by 1.93.95.8+1.9
ReasoningCeleris-1
- GPQA DiamondCeleris-1 by 19.363.143.8+19.3
- Humanity's Last ExamCeleris-1 by 2.36.84.5+2.3
- CritPttie00tie
FactsEven
- AA-Omniscience · Non-hallucinationG9v3-3B by 81.17.288.3+81.1
- AA-Omniscience · AccuracyCeleris-1 by 4.5116.5+4.5
Other results3 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.
- AA-OmniscienceG9v3-3B by 67.2-71.6-4.4+67.2
- AA IntelligenceG9v3-3B by 4.56.310.8+4.5
- Artificial Analysis Coding IndexCeleris-1 by 4.514.49.9+4.5
Questions people ask
Which is better, Celeris-1 or G9v3-3B?
Celeris-1 and G9v3-3B each win two of the five areas where both have results. Celeris-1 wins coding and reasoning; G9v3-3B wins agents and long documents. They are level on facts.
Which is better for coding?
Celeris-1. It wins 2 of the 2 coding tests both models report; G9v3-3B wins none.
How do you compare the two?
We use the 13 benchmark tests both models have published scores on. The verdict counts the 10 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 3 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.