Will verified AI inference performance per watt double by October 2027?
MLPerf Endpoints: at least 2x GPT-OSS-120B standardized throughput per watt. The question has a 27 October 2027 deadline, but no official Cortist probability or scored forecast vintage exists yet.
Evidence Pulse 2.0 checks the published Cortist News index when JavaScript is available. The source-linked, dated research articles remain accessible through Cortist News.
- Question ID
- CM-005
- Deadline
- 27 October 2027, 23:59 UTC for observable events
- Outcome measure
- MLPerf Endpoints: at least 2x GPT-OSS-120B standardized throughput per watt
- Forecast status
- Candidate definition only · no probability · not locked · not scoreable
- Evidence vintage
- First candidate text published 27 October 2026
What do we know today?
MLCommons launched continuously reviewed Endpoints results in 2026; cross-vintage power normalization still requires a stable baseline.
Chips and inference efficiency need full-system like-for-like measurement.
How would this milestone be resolved?
What qualifies as YES
MLCommons verifies an Available, Standardized GPT-OSS-120B system with at least twice the pre-registered 2026 reference output tokens per watt at matching latency and quality.
What qualifies as NO
Official comparable results remain below the locked 2x threshold at cutoff.
What would not qualify?
Different models, changed quality limits, vendor-only figures and Preview/RDI submissions do not count.
Unresolved definition gate
Critical: freeze 2026 baseline, comparable power normalization, model/tokenizer, quality tier, latency and reviewer-approved rules.
Consequence: no probability should be locked or scored before this issue is resolved and the protocol is externally timestamped.
Why does it matter?
Chips and inference efficiency need full-system like-for-like measurement. MLCommons launched continuously reviewed Endpoints results in 2026; cross-vintage power normalization still requires a stable baseline.
Interpretation limit: Chips and inference efficiency need full-system like-for-like measurement. An individual benchmark or adoption result is not proof of AGI or a universal model capability.
Primary sources and verification references
- Primary source / methodology (original reference)
- Official reference / evidence (original reference)
What will Cortist publish next?
Before treating this candidate as a numerical forecast, Cortist must freeze the evaluation protocol, verify that the event is not already resolved, define an independent public evidence source and record a probability in a new, immutable version. The present candidate page is not part of the scored probability history.
Read Cortist’s public Forecast Protocol · View the existing immutable forecast register · Explore the AI Chips & Compute Topic.
Machine-readable candidate record: v1 JSON. Original candidate publication: 27 October 2026.