Nicheloom

Market intelligence for builders — see what's gaining traction before it's crowded.

Optimize and serve models with Fable quality at half the cost

Details

External ID
49063454
Source
HN
Company
—
Product
Optimize and serve models with Fable quality at half the cost
Website domain
github.com
Launched
July 26, 2026
Cohort
—
Upvotes
71
Upvotes percentile
0.8823178016726404
Tags
—
Fetched at
Sept. 7, 2026, 9:26 p.m.
Updated at
Sept. 7, 2026, 9:26 p.m.

Description

Hi HN, we built world-model-optimizer, an open source tool to continually improve a specialized model for an agent.It does this by simulating production tool responses through text world modeling (similar to QwenAgentWorld, summary here https://x.com/silennai/status/2073887455884058814).We can then use this to train a router for frontier, OS, and local models (use defaults or pick which ones to optimize against).wmo ingests agent traces, builds the simulation, embeds the traces, runs different models you choose against the simulation scenarios, and then uses a KNN for model selection (similar to https://arxiv.org/abs/2505.19797).- Cache aware: cache is taken into account for the effective price in routing.- Confidence gated: we don't deviate from the best fit model when paired evidence over retrieved neighbors is below 0.5 standard errors or on queries unlike anything in the fit set.- Optimize for cost or quality: train a balanced, cost max, or quality max router.Usage`wmo build` creates the simulation (or add your own benchmark)`wmo optimize` tunes the router`wmo serve` starts the server and can run everything fully locally. The simulation and router can update over time as more agent traces are gathered and new models are added.Router results vs Fable- RouterBench: -66.5% cost, -1.7% performance, -24.7% latency p50. 77.5% of traffic to Sonnet 5, 16.1% Fable 5.- TauBench: -44.5% cost, +6.3% performance, -20% latency. 83% to Opus 5, 17% to Kimi-K2.6 (over K3).- Terminal Bench 2: -64% cost, +8% performance, -50.6% latency. Sonnet 5 is fully along the pareto front. Training a specialized router per task isn't cheap. In sparse data regimes the value can be "here's the best model".We're working on sample effiient continual learning for agent specific models at experientiallabs.ai"

Enrichment

Theme
lightweight and on-device AI runtimes
Vertical
Horizontal
Function
Model & infra
Audience
B2B
AI stance
AI-native
Project type
Commercial product
Normalized one-liner
optimize and serve ai models at lower cost
Manually corrected
False

Could you build this?

No Building an LLM world-model optimizer that simulates dynamic production tool environments to fine-tune and distill agent models involves cutting-edge AI research and distributed model training.

What it would actually take: The architecture requires PyTorch, vLLM/DeepSpeed, synthetic data generation pipelines, and reinforcement learning / rejection sampling fine-tuning frameworks (DPO/PPO). The hard part is training an accurate text world-model that simulates complex multi-step tool call responses realistically and using that environment to distill or align agent models without catastrophic drift. This demands machine learning research scientists with deep experience in agent trajectories, synthetic data curation, and GPU cluster training.

Discussion

20 comments analyzed.

Competitors mentioned: Frontier models (GPT-4, Claude, etc.) for task routing, Other distillation approaches

Concerns raised: Simulator exploitation/reward hacking in isolated environments, High training costs (thousands of dollars), Unclear cold-start performance with limited traces, Reconstruction fidelity vs. downstream performance disagreement, Results not yet solidified/mature enough

Feature requests: Pre-trained routers in app for faster onboarding, Cost reduction for training and inference, Better documentation on traces needed for reliable routing

Competitors

Other products that read as similar to this one — 67 launches clear the similarity bar, closest 8 shown.

Attention rank: #10 of 68 (itself plus its competitors, highest first — normalized so YC and Product Hunt are compared fairly).

Launched 270 days after the earliest competitor.

Other launches for this product

Same idea, different domain

Nobody's really built a model & infra tool for Fintech yet.