Latest model · #1 on GTMBench
Nico2.8

The #1 model on GTMBench — built to act, not just answer.

Nico 2.8 is Macrodeep’s latest model — a large language model trained end-to-end for go-to-market: research, outreach, qualification, and the judgment that ties them together. It powers every Macrodeep product — and it’s available to build on directly through the API.

NNico 2.8nico-2.8
Built forGo-to-market work
GTMBench#1 overall
BehaviourAgentic by default
Context window500K tokens
Tool use75+ native connectors
OutputsText · structured · tool calls
In productionNico · Salestools · Grain
ReleaseNico 2.8 · 2026

Most models describe the motion. Nico 2.8 runs it.

A general model can write an email. A GTM model knows which account, why now, and what to do when the reply lands. Nico 2.8 is trained on the full loop — sourcing, enrichment, scoring, sequencing, and qualification — so it reasons in the currency of pipeline, not prose.

01What it’s built for
01

Runs the whole motion

plan → enrich → score → sync

Hand Nico 2.8 a goal and it plans the run, calls the tools, and drives it to a finished state — not a paragraph of advice, but accounts sourced, scored, and synced.

02

Reads intent, not keywords

judgment on every reply

It scores replies, buying signals, and ICP fit the way a strong rep would — weighing context, timing, and history to decide what deserves a human and what it can advance.

03

Tool-native by default

75+ connectors in the loop

Salesforce, HubSpot, Clay, Slack and 70+ more aren't plugins bolted on after training — Nico 2.8 learned to call them, chain them, and recover when one fails.

02Benchmarks

A GTM model, measured on GTM work.

Higher is better · GTMBench score, 0–100 · September 2026 run
Nico 2.8Nico 2.5GPT-6 Astra
Overallcomposite of all eight families
93.2
88.4
89.6
Account researchmulti-source intel synthesis
92.2
87.7
92.9
Company & persona researchcited dossier on the buying committee
93.1
88.6
94.7
Outbound copyspecificity and deliverability
96.3
92.9
82.5
Call analysistranscript to qualification and next steps
89.3
84.7
94.2
Objection handlingmulti-turn negotiation
96.4
91.2
87.6
CRM extractionschema-strict extraction and dedup
90.3
85.7
89.4
Sequence executionagentic multi-step runs with tool calls
92.5
85.0
87.1
End-to-end account planoutreach plan across personas and channels
95.5
91.5
88.4

Scores from GTMBench, the public go-to-market benchmark — September 2026 run, 2,946 graded tasks across 30 models. Nico 2.8 ranks #1 overall at 93.2; the next-best model, GPT-6 Astra, scores 89.6 and leads on account research, company & persona research, and call analysis.

03Availability & pricing

Everywhere Macrodeep works — including your code.

01Nico — the GTM engineer app on macOS, iPhone, and iPad.
02Salestools & Grain — the model behind every autonomous agent.
03Frontier — deployed inside the enterprise, tuned to your domain.
04The API — one keyless endpoint, metered per token. Build on it →
ModelInputOutput
Nico 2.8nico-2.8
$6 / M tok
$30 / M tok

Illustrative list pricing per million tokens. Tool calls to connectors are metered separately.

04Capable and controllable

Autonomy you can actually deploy.

A model that acts needs oversight built into the weights, not bolted on with a prompt. Nico 2.8 is trained to respect the approval boundary — and every run is legible, interruptible, and logged.

01Approval boundary. Sends, writes, and money-moves are held for a human — enforced at the tool boundary, not requested in a prompt.
02Legible reasoning. Every draft carries the signal, confidence, and persona behind it, so you review with context.
03Full audit trail. Every step and tool call is logged and replayable. Interruptible mid-run, always.
04No training on your data. Your accounts, replies, and outcomes never train shared models.

Build on the model that acts.

API access is coming soon. Try Nico, the app, today.