Moonshot Releases Open-Source Kimi K3 with 2.8T Parameters, Inference Hits 423 tok/s on Day One
Moonshot officially released the weights and technical report of its open-source flagship model Kimi K3. The model has 2.8T total parameters, uses a MoE architecture with 104B active parameters per token, supports a 1M-token context window, and has native vision understanding. It emphasizes efficient compute utilization rather than brute-force scaling, and has received rapid inference optimization support from the open-source community and multiple platforms on day one.
Confirmed
- Architecture: Kimi K3 has 2.8T total parameters, 104B active parameters, native vision, and 1M context. It focuses on efficient compute use. According to @noninertialframe96's interpretation of the technical report, architecture highlights include Delta Attention, expert balancing, and AgentENV.
- Benchmarks: It is rated the strongest open-source model on the Public Intelligence Index and ranks fourth on Artificial Analysis, behind Claude Opus 5, Fable 5, and GPT-5.6 Sol.
- Platform support: The model is available on Modal, Featherless, and other platforms.
- Inference optimization: SGLang achieved 423 tok/s on GSM8K on day one, with native RL support, thanks to community and NVIDIA/AMD operator optimizations. Modal also provides 460 TPS.
Why it matters
- Kimi K3 is one of the largest open-source models (nearly 3T parameters), marking a step closer to top closed-source models in scale and capability.
- The immediate high-speed inference support (e.g., SGLang >400 tok/s) demonstrates the open-source ecosystem's rapid response in software-hardware co-design, lowering deployment barriers for ultra-large models.
2026-07-29 ~ 2026-07-31 · 7 related posts
- Episode 1: Kimi K3 Tops Frontend Web App Arena with Enhanced English Skills(2026-07-21, 3 posts)
- Episode 2: Kimi K3 Jumps to 4th on Agent Arena Leaderboard(2026-07-21, 6 posts)
- Episode 3: Moonshot Releases 2.8 Trillion Parameter Open-Weight Model Kimi K3(2026-07-21, 8 posts)
- Episode 4: Kimi K3 Matches Fable 5 in SWE Benchmarks at a Third of the Cost(2026-07-21, 7 posts)
- Episode 5: Kimi K3 Sets New Open-Source ECI Record but Still Lags Behind(2026-07-22, 3 posts)
- Episode 6: Kimi K3 Accused of Gaming Benchmarks Instead of Solving Problems(2026-07-22, 2 posts)
- Episode 7: Kimi K3 Ranks Second on AA-Briefcase but with High Costs and Long Runtimes(2026-07-22, 6 posts)
- Episode 8: Kimi K3 Enters Top-Tier AI Model Ranks in Benchmark Tests(2026-07-22, 4 posts)
- Episode 9: Kimi K3 shifts attention from scale to architecture(2026-07-27, 25 posts)
- Episode 10: TokenSpeed Enables Kimi K3 Support on NVIDIA and AMD Platforms(2026-07-27, 2 posts)
- Episode 11: Moonshot's Kimi K3 Launches on Nebius with 1M Context(2026-07-27, 3 posts)
- Episode 12: SGLang Day-0 Support for Kimi K3 Boosts Throughput to 423 tok/s(2026-07-28, 8 posts)
- Episode 13: Moonshot's Kimi K3 Flagship Model Launches on Together AI(2026-07-28, 12 posts)
- Episode 14: Kimi K3 Max Tops Multiple Arena Leaderboards, Open-Source Model Rivals Proprietary(2026-07-28, 11 posts)
- Episode 15: Fireworks Test: Kimi K3 Matches Opus 5 Quality at Fraction of Cost(2026-07-28, 5 posts)
- Episode 16: Kimi K3 Impresses in Early Benchmarks, Sparking Buzz(2026-07-28, 3 posts)
- Episode 17: Local Kimi K3 Beats Cloud Models in 3D Physics Generation Test(2026-07-28, 5 posts)
- Episode 18: Deep Dive into Kimi K3 Tech Report: Engineering Synergy Drives State-of-the-Art Performance(2026-07-28, 21 posts)
- Episode 19: Moonshot AI's Kimi K3 Launches in Japan(2026-07-28, 2 posts)
- Episode 20: Kimi K3 Passes Compound Benchmark Amid Cost Efficiency Concerns(2026-07-29, 2 posts)
Primary sources
- Moonshot Releases 2.8T-Parameter Kimi K3; Modal Achieves 460 TPS with Speculative Decoding — sarahcat21 ·
- Moonshot’s Kimi K3 is a 2.8T open-weight MoE model with 1M-token context — alex_verem ·
- How Kimi K3 engineered its way to the frontier: Delta Attention, expert balancing, AgentENV — noninertialframe96 ·
- [source] Moonshot’s Kimi K3 is a 2.8T open-weight MoE model with 1M-token context — alex_verem · 2026-07-29
- [source] Moonshot Releases 2.8T-Parameter Kimi K3; Modal Achieves 460 TPS with Speculative Decoding — sarahcat21 · 2026-07-30
- SGLang Ecosystem Relay: Kimi K3 Hits 423 tok/s on Day-0 with Deep Optimizations — songhan_mit · 2026-07-30
- Kimi K3 Released: 2.8T Open-Weight Flagship with Native Vision and 1M Context — burny_tech · 2026-07-30
- Kimi K3 Released: Nearly 3-Trillion-Parameter Open Model Rivals Closed-Source Frontier — Aizkmusic · 2026-07-30
- Kimi K3 Arrives as Largest Open-Source Model Amid a Week of AI Open Letters — altryne · 2026-07-30
- [source] How Kimi K3 engineered its way to the frontier: Delta Attention, expert balancing, AgentENV — noninertialframe96 · 2026-07-31