AI News/Wire summary
Gemini 4 Carbon leads MathArena, at roughly 15 times GPT-6.1 Sol's cost
Key points
- Carbon knocked GPT-6.1 Sol off the top of the MathArena math benchmark, with Claude Fable 5.2 placing second.
- Carbon most often fails by running past the time limit, and it rarely gives up, once running a script over eleven hours unchecked.
- Under the reasoning settings used in testing, Fable 5.2 reportedly costs about 40 times as much as GPT-6.1 Sol.
- According to the report, OpenAI models allegedly solved Navier-Stokes equations, but those models are not publicly available yet.
Why it matters for marketers: Teams choosing models for math-heavy features should weigh the benchmark gain against much higher per-task cost and the risk of runaway runtimes.
Summary by the Z Media news desk, from the outlets' reporting. Read the full stories at the source.
Read the original
The Decoder: Gemini 4 Carbon takes the MathArena crown but costs 15 times more than GPT-6.1 Sol
Related on Z Media
- Google adds Gemini 4 Neon as a higher-tier model for agent and coding workOur brief, via The Decoder, 2 days ago
- OpenAI's GPT-6.2 Astra and Sol cut costs, but outside tests show mixed resultsOur brief, via OpenAI, 3 days ago
- Gemini for Mac can now operate your apps and files on its ownOur brief, via TechCrunch, yesterday
- Anthropic prices IPO at $83 a share, raising a record $81.3 billionOur brief, via Anthropic, 2 days ago
- Nvidia reportedly nears $500M deal for diffusion model startup Inceptionvia The Decoder, 9 hr ago
- Anthropic CEO pledges to pace AI development at Nasdaq IPO debutvia TechCrunch, 2 days ago