Fireworks' official launch post (published 7/27/2026) confirms that Kimi K3 is available Day-0 on Fireworks for both inference and training as an open-weight model, with US-hosted inference and zero data retention aimed at regulated industries. The post describes Kimi Delta Attention as enabling faster tokens at lower The launch framing positions open models as having "crossed the line" to match closed frontier quality at a fraction of the cost, encouraging teams to default to an open model like Kimi K3 and only route to specialized intelligence where tasks demand it. It emphasizes that training a 3T-class model can start from a lap

The Fireworks blog index lists Kimi K3–related posts spanning late July through August 26, 2026, including the model's Day-0 launch ("Kimi K3 on Fireworks: Frontier Intelligence You Can Own," 7/27/2026), a LoRA training guide ("Make Kimi K3 Yours: LoRA Training on Fireworks," 7/26/2026), a partner post-training case st The same index shows broader model-release cadence around Kimi K3, including DeepSeek V4 Pro topping SWE-Bench with 3x lower cost per task (8/26/2026), Trilogy's open-weight cybersecurity playbook with Kimi K3 (7/26/2026), and the claim that "Kimi K3 is competitive with Fable; Kimi K3 + Fable is SoTA" (7/21/2026). Earl

GitHub announced on August 6, 2026 that Kimi K3 is generally available in GitHub Copilot, hosted by GitHub on Fireworks AI under usage-based billing at provider list pricing. The post documents a brief rollout pause due to a GitHub Actions incident that was later resumed the same day, with final pricing set at $3 per 1 Business and Enterprise administrators must explicitly enable the Kimi K3 policy in Copilot settings; if left off, the model remains unavailable to the organization, and GitHub recommends that admins review open-weight models against their own security, compliance, and data-governance requirements before enabling. The

A July 21, 2026 benchmark from Fireworks AI compared Moonshot's open Kimi K3 model against Anthropic's Fable 5 on roughly 1,030 agent-based software development tasks, scoring Kimi K3 at 92.4% versus Fable 5 at 92.6%, placing the two within a fraction of a point of each other. On long-running agent processes the report For developers evaluating Fireworks AI's hosted Kimi K3 endpoint, the comparison implies near-parity agent coding performance at a substantially lower token cost for sustained workflows, though the underlying Fireworks benchmark page itself was not surfaced in the scraping set and is referenced only second-hand. The re

Simon Willison's July 16, 2026 post documents Moonshot AI's Kimi K3 launch as a 2.8-trillion-parameter model, positioned as the first "open 3T-class" release, with open weights promised by July 27, 2026 and immediate availability via Moonshot's website, API, and consumer products. Moonshot's self-reported benchmarks pl For developers using Fireworks AI's OpenAI-compatible endpoint, the post confirms relevant pricing context at $3 per million input tokens and $15 per million output tokens, matching Anthropic's Sonnet tier and making Kimi K3 the most expensive Chinese-lab release to date (up from $0.95/$4 for K2.6). Willison's hands-on