MAI Code 2 Flash
AvailableSmaller, low-cost sibling of MAI Code 2, launched Nov 17 2026: generally available in GitHub Copilot (including Copilot Free) and in limited-access preview in Microsoft Foundry. Sparse MoE with vision encoder, trained from scratch with logit distillation from a larger MAI teacher; text and image input, 1,050,000-token context (922K input), 128K max output. Model card detail gives about 398B total / 12B active parameters, while its summary table says 5B-15B (12B active), so size fields are left blank. Copilot price $0.10 / $0.50 per 1M input/output tokens, with a 50% promo ($0.05 / $0.25) until May 17 2027. Succeeds MAI-Code-1.1-Flash. Vendor-reported benchmarks.
Specifications
- License
- Proprietary
- Weights
- Not released
- Architecture
- Mixture-of-Experts
- Parameters
- Undisclosed
- Context window
- 1.1M tokens
- Max output
- 128K tokens
- Knowledge cutoff
- —
- Price (in / out, $/M)
- $0.1 / $0.5
- Modalities
- TextVisionCode
Benchmarks
No benchmark scores recorded yet. Spotted some? Submit a correction.
Vendor-reported figures are claims until independently verified. See methodology.