
By what factor will the cost for SotA SWE-agents drop from 2024 to 2025?
9
1kṀ871Jul 2
5%
<2x
8%
<10x
9%
<50x
21%
<250x
57%
>=250x
Algorithmic progress can be measured by reduction in cost to achieve equivalent performance. SWE-bench-lite is a popular benchmark for measuring scaffolded-LLM SWE capabilities.
By what factor will the cost of SWE-bench-lite SoTA drop between mid 2024-2025? Mid-2024 SotA is 43% costing $2,700 (per the devs), so this question will resolve Yes on the answer which most tightly bounds the reduction in cost to achieve 43% on July 1, 2025.
E.g. if in June 2025, 43% on SWE-lite costs $500 then that'd be a 5.4x reduction and the question would resolve (2) "<10x".
This question is managed and resolved by Manifold.
Get
1,000 to start trading!
People are also trading
Related questions
Will some U.S. software engineers be negatively affected financially due to AI by end of 2025?
99% chance
If AI does not wipe us out by 2033, will there be continued progress in the AI sector?
79% chance
Will AI be Recursively Self Improving by mid 2026?
38% chance
Will an autonomous agent resolve 90% of tasks on SWE-bench by 2026?
63% chance
How much will AI advances impact EA research effectiveness, by 2030?
Will an autonomous agent resolve 90% of tasks on SWE-bench by 2027?
72% chance
Will I believe my prediction about AI enabling more SWEs to solve less lucrative problems by shrinking team sizes to have been fulfilled by EoY 2030
68% chance
What will AI score on TheAgentCompany benchmark in early 2026?
42% chance
Will OpenAI release an agent priced at $10k per month or above in 2025
7% chance
Will we reach "weak AGI" by the end of 2025?
25% chance