Frontier LLM Registry

⚠ US-GOV HOLD — temporarily blocked in the U.S. by federal order. Access restricted pending review.

First Claude 5 model. Mythos-class tier above Opus; most intelligent generally available Claude. Added dual-use safety measures vs. the Mythos sibling.

Final Opus 4.x point release; faster output in Claude Code fast mode; stronger agentic tool use.

Real-time X integration; agent improvements; faster tool loops.

Unified reasoning model; auto-router replaces o-series picker.

2M token context; native video understanding; Deep Think mode.

High-throughput tier; multimodal at low latency.

On-device tier; ships with newer Pixel/Android.

Long-context reasoning improvements; better tool use under uncertainty; reduced sycophancy. Superseded by Opus 4.8 / Fable 5.

Open-source agent model; trained on real tool-call traces.

Search-grounded; web-native answers; citations first-class.

Speed/cost-optimized; near-Opus reasoning on most benchmarks.

Chain-of-thought reasoning successor to R1; open weights.

RAG-specialized; tool-use leader at this tier.

MoE architecture refinements; aggressive pricing remains.

Agent-tuned coder model; open weights.

Sub-second latency; agent workflows; budget tier.

Improved multilingual; function calling parity with frontier.

Massive multilingual; strong code; competitive with frontier closed models.

2T total params; teacher model; not open-weight (yet).

Initial GPT-5 launch; replaced o3/o4-mini routing.

Multi-agent reasoning tier (premium plus).

Bilingual leader (en/zh); strong reasoning at mid-tier price.

MoE; 17B active / 400B total; instruction-tuned flagship open weights.

10M token context; MoE 17B active / 109B total.

Reasoning-tuned mini; superseded by GPT-5 auto-router.

Code-specialized; 80+ languages; FIM optimized.