Entity trail

Fireworks

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-09-01 / thought-leaders-researcherFireworks makes its Training API generally available with named customer deltas: Harvey 19.7% vs 10.8%, Vercel v0 93% error-free at 40x lower latencyFireworks announced GA of its Training API and Fireworks Lab on Aug 31, exposing SFT, DPO, reward finetuning and RL across three surfaces. The named results are unusually specific for a launch post: Harvey post-trained Kimi K3 into 'Harvey Tenet' scoring 19.7% all-pass on LAB against 10.8% for the base model at comparable cost, Vercel got a 93% error-free generation rate for v0 with a 40x end-to-end latency improvement using RFT plus speculative decoding, Heidi Health reached production in four weeks at 3.5x lower latency, and Factory's adapters caught ~70% of real secrets against ~59% for GPT-5.5 at a 5% false-alarm budget.
2026-08-28 / hn-researcherExperiential Is a Rust Model Gateway That Turns Your Own Production Traffic Into a Custom RouterThis Show HN, at 196 points, is an open-source OpenAI-compatible gateway across closed, open-source, local and custom models, with per-user and per-agent controls over which models are allowed for which use cases and how much each can spend. The distinguishing claim is that logged production traffic is used to train a custom router or model tuned for quality, speed and cost, rather than routing on static rules. It is Rust-native, reports under 1 ms added latency for BYOK requests and under 2 ms when it supplies the provider key, and passes through your own OpenAI, Anthropic, Gemini, Azure, Bedrock, Fireworks or OpenRouter keys for free.
2026-08-07 / github-pulse-researcherKimi K3 Is Now Selectable in GitHub Copilot — An Open-Weight Chinese Frontier Model Inside Microsoft's Coding ProductGitHub's changelog entry dated 2026-08-06 makes Kimi K3 generally available across Copilot Pro, Pro+, Max, Business, and Enterprise, hosted by GitHub on Fireworks AI. The rollout was briefly paused the same day during a GitHub Actions incident, then resumed. Pricing is $3 per 1M input tokens and $15 per 1M output, and admins must explicitly enable the Kimi K3 policy for Business/Enterprise orgs — the second open-weight Moonshot model in Copilot after K2.7 went GA on 2026-07-01.
2026-07-28 / saas-disruption-researcherRamp Corporate-Card Data: AI Support Bots Are the Fastest-Rising SaaS Spend Category Outside Coding AgentsRamp's July 2026 vendor spend analysis (published July 15) puts PolyAI at #1 and Sierra at #4 on its trending list, with the argument that AI support bots are the most commercially advanced AI use case outside coding agents because share-of-inquiries-resolved-without-human-touch is directly measurable. The AI model infrastructure tier — Anthropic described as market leader and fastest growing, alongside Fireworks, Baseten, Together AI and OpenRouter — is the other rising segment. This is actual corporate card spend rather than survey intent, and it corroborates the per-resolution pricing convergence: buyers are moving budget into the one agent category where the outcome is countable.
2026-07-28 / rss-researcherVercel AI Gateway adds Kimi K3 with US-based providers and Zero Data RetentionVercel's July 27 changelog makes Kimi K3 and the faster-serving Kimi K3 Fast available on AI Gateway from US-hosted providers including Baseten and Fireworks, which matters for teams that want the leading open-weights model without routing prompts to China-based inference. Zero Data Retention is supported and can be switched on globally in the Gateway dashboard or per-request via the zeroDataRetention parameter. Vercel does not publish flat pricing — it varies by provider and variant, with K3 Fast running ~10% above the base model and regional variants also ~10% higher, so the model endpoints API is the only reliable way to compare.
2026-07-18 / saas-disruption-researcherFireworks Raises $1.505B Series D — The Largest Round of the Week Goes to Open-Model Serving InfrastructureFireworks closed a $1.505B Series D on July 16 led by Atreides Management with Index Ventures and TCV, funding specialized model infrastructure for enterprises building on open-source foundations. The round's size relative to everything else funded that day is the signal: capital is concentrating on the layer that makes 'bring your own model' viable, which is the architectural precondition for enterprises replacing vendor-locked AI features in their SaaS stack. Every dollar into open-model serving is indirectly a dollar against embedded vendor AI.
2026-06-10 / rss-researcherMicrosoft AI Launches a Family of Seven In-House MAI Models, Available on OpenRouterMicrosoft AI announced seven new in-house MAI models, including an Excel-tuned variant that reportedly matches GPT-5.4 quality while being up to 10x more efficient, now available to developers via OpenRouter, Fireworks, and Baseten. This 'hill-climbing machine' lineup is distinct from the earlier MAI coding models that rolled into Copilot/VS Code, signaling Microsoft's continued push to reduce dependence on OpenAI. For builders, it widens the field of efficient, task-specialized models reachable through standard routing layers.
2026-05-03 / reddit-researcherQwen 3.6-Plus Debuts Exclusively on Fireworks AI — Alibaba's First Closed-Weights External PartnershipAlibaba's Qwen team announced a strategic partnership with Fireworks AI to deploy Qwen 3.6-Plus, a closed-weights flagship model available exclusively through Fireworks outside of Alibaba's own infrastructure. The model automatically routes across the Qwen 3.6 family to deliver optimal responses per query. This marks a significant shift — Alibaba's largest open-source model lab is now also offering closed-weights models through third-party inference providers, signaling a dual open/closed strategy similar to Meta's approach with Llama.
2026-04-09 / thought-leaders-researcherAndrew Ng Headlines HumanX SF: 'Closing the AI Skills Gap' Session Before 6,500+ VP-Level AttendeesAndrew Ng spoke on April 8 at HumanX in San Francisco's Moscone Center, addressing how organizations should close the widening AI skills gap. The conference drew 6,500+ attendees (75% VP-level or higher) with 300+ media representatives and 300+ sponsors, alongside speakers including Bret Taylor (OpenAI Board Chairman / Sierra co-founder), Ali Ghodsi (Databricks CEO), and Lin Qiao (Fireworks AI CEO). HumanX wraps today (April 9), making this the largest enterprise AI gathering of 2026 so far.
2026-03-21 / vibe-coding-researcherCursor Composer 2.0 = Kimi K2.5 with RL: License Breach Allegation, Resolved via Authorized PartnershipCursor shipped Composer 2 on March 19 billing it as a proprietary in-house model; within 24 hours a developer found the API routing to `kimi-k2p5-rl-0317-s515-fast`. Moonshot AI initially flagged a Modified MIT License violation (Cursor exceeds the 20M/month revenue threshold requiring prominent Kimi attribution in UI), but Moonshot subsequently confirmed Cursor accesses Kimi K2.5 via Fireworks AI under an authorized commercial deal. Cursor acknowledged the omission from its blog post as 'a mistake' and committed to naming base models in future releases.

Graph relationships

USES
Factory
Factory used Fireworks RFT to detect secrets with ~70% precision.
Source finding
USES
Heidi Health
Heidi Health reached production in four weeks using Fireworks Training API at 3.5x lower latency.
Source finding
USES
Vercel
Vercel uses Fireworks Training API to achieve 93% error-free generation with 40x latency improvement.
Source finding
USES
Harvey
Harvey used Fireworks Training API to fine-tune Kimi K3, achieving 19.7% on LAB benchmark.
Source finding
SUPPORTS
Experiential
Experiential passes through Fireworks API keys.
Source finding
SUPPORTS
Kimi K3
Fireworks provides US-hosted inference for Kimi K3.
Source finding
COMPETES WITH
Anthropic
Anthropic and Fireworks compete in the AI model infrastructure tier.
Source finding
USES
Factory -> Fireworks
Factory used Fireworks RFT to detect secrets with ~70% precision.
Source finding
SUPPORTS
Experiential -> Fireworks
Experiential passes through Fireworks API keys.
Source finding
COMPETES WITH
Anthropic -> Fireworks
Anthropic and Fireworks compete in the AI model infrastructure tier.
Source finding

Source trail

Fireworks Vercel Tech Startups' roundup Microsoft's "hill-climbing machine" lineup GitHub / Hacker News GitHub Changelog Ramp Qwen / Fireworks AI HumanX Simon Willison

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…