Llama 4 Maverick 17B 128E Instruct FP8: documented context capacity is one million tokens
Meta's new Llama 4 17B variant achieves a massive million-token context capacity for enhanced efficiency.
Documented context capacity is one million tokens — verified by the OnAirToday AI Progress Tracker against Meta AI's published documentation.
OnAirToday
Newsroom
The OnAirToday AI Progress Tracker recorded a verified change to Meta AI's Llama 4 Maverick 17B 128E Instruct FP8 (version Llama-4-Maverick-17B-128E-Instruct-FP8): Documented context capacity is one million tokens.
Meta's FP8 instruction-tuned Llama 4 Maverick is a multimodal mixture-of-experts model with 17 billion active parameters and 128 experts. It accepts text and images, generates text, and has a documented one-million-token context window; hosted inference pricing depends on the provider.
What changed in this review:
- Meta reports Maverick GPQA Diamond at 69.8%; this is distinct from the previously listed, unverified Epoch result.
- Added published Maverick MMLU-Pro, LiveCodeBench and MMMU results.
- Unverified OTIS Mock AIME, FrontierMath and MATH Level 5 figures are not retained as established results.
- Pricing remains unspecified: downloadable weights do not establish free hosted inference or a universal API rate.
- Capability score 43 → 67
- Flaws and limitations updated
Teams building on Llama 4 Maverick 17B 128E Instruct FP8 should confirm limits, pricing and benchmark coverage against the vendor's own documentation before relying on these figures.
The full profile — capabilities, benchmark citations, known flaws and risk guidance — is on the tracker at /ai-models/llama-4-maverick-17b-128e-instruct-fp8.
References:
- https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8
- https://ai.meta.com/blog/llama-4-multimodal-intelligence/
- https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8/blob/main/LICENSE
Generated insight by OnAirToday AI
The introduction of a one-million-token context window in the Llama 4 17B model represents a major leap in efficiency for mid-sized LLMs. This capability allows developers to process massive datasets locally without sacrificing the performance advantages typically reserved for much larger, resource-heavy architectures.

Comments
Keep comments relevant and respectful. No links, advertising, self-promotion, or spam.