Nine models. Five countries. Four continents.

Flagship reasoning model from OpenAI. Strong on long-form synthesis and source attribution.

Documented left-of-centre political tilt; cautious on contested empirical claims.

High-context reasoning model. Known for explicit uncertainty calibration in long answers.

Tends toward centrist framing; conservative on confidence claims.

Counterweight model with deliberately non-default training posture; surfaces contrarian framings.

Less left-leaning than US peers but still empirically not right-of-centre.

Strong multilingual coverage; particularly capable on geographic and scientific claims.

Centrist political profile; sometimes hits output caps on long claims.

European editorial counterweight; trained on a corpus with stronger EU regulatory framing.

Moderate left tilt comparable to US peers; GDPR-aware in privacy claims.

Strong reasoning model; provides non-Western training perspective.

State-aligned framing or refusals on PRC-sensitive topics; captured as data.

Best CJK and Arabic-language coverage in the panel; useful on East Asian claims.

Same state-alignment patterns as DeepSeek; different training lineage.

South-Asian perspective; eleven Indic languages plus English.

Mistral-derived base; Indic post-training. Not a fully independent training lineage.

Arabic-first training corpus; provides Gulf editorial perspective.

Cautious on Gulf-state political topics; otherwise unaligned in Western political terms.