Sign InOpen Brain
AI EngineerVideoSource Linked

The Base Model Is Dead — Varun Singh, Arcee AI

Base-model data is shifting from broad web imitation toward code, reasoning, and agent-task priors. The unresolved choice is how early to introduce synthetic and instruction-shaped data.

AI Engineer · Jul 31, 2026
Open Source Open MarkdownOpen JSON
Source Summary

The talk contrasts GPT-3’s roughly **85% web-derived mix** with MAI Thinking 1 at **15% web text**. Newer recipes emphasize code, STEM, reasoning traces, and task-shaped data that better prepare models for downstream RL.

Practical Implication

When selecting or training a model for agents, evaluate its pre-RL skill coverage, not just general knowledge. **NeMoTron 3 Ultra** pulls SFT-style data into pre-training, while synthetic rephrasing can expose the same information in several forms.

Agent-Ready Context
The talk contrasts GPT-3’s roughly **85% web-derived mix** with MAI Thinking 1 at **15% web text**. Newer recipes emphasize code, STEM, reasoning traces, and task-shaped data that better prepare models for downstream RL.

When selecting or training a model for agents, evaluate its pre-RL skill coverage, not just general knowledge. **NeMoTron 3 Ultra** pulls SFT-style data into pre-training, while synthetic rephrasing can expose the same information in several forms.

There is no settled recipe: MAI Thinking 1 deliberately avoids synthetic model-generated data, while NeMoTron leans into it. Synthetic data can degrade a model when used indiscriminately, and it remains unclear how far RL can displace supervised learning for language.
Connected Context · Feed7 Judgment

This shifts model selection beneath gateway features and headline benchmarks to the capabilities and data composition established before RL. It explains why nominally general models may differ sharply on agent work, while refusing a universal recipe: synthetic data, SFT-like pre-training, and RL remain choices whose value must be established on the intended workload rather than inferred from model branding or context size.

Introducing Grok 4.5Grok 4.5’s contaminated coding benchmark reinforces the target’s demand to inspect training provenance and pre-RL capability evidence rather than accept a headline score at face value.DeepSeek V4 Flash now runs updated weights on AI GatewayThe same model ID receiving materially different weights shows the implementation consequence of recipe-dependent capability: selection requires continuous workload evaluation even when routing configuration is unchanged.Inkling Small from Thinking Machines is now available on AI GatewayInkling Small’s efficiency and adjustable-effort positioning supplies a routing candidate, but the target narrows selection to whether its underlying skill coverage matches the agent workload, which the candidate’s absent comparisons do not establish.
Context Map
modelcoding#reasoning#coding-agents#model-selection
Uncertainty
There is no settled recipe: MAI Thinking 1 deliberately avoids synthetic model-generated data, while NeMoTron leans into it. Synthetic data can degrade a model when used indiscriminately, and it remains unclear how far RL can displace supervised learning for language.