Rankings are easy to publish and hard to trust. The October Fast Model Index is the first monthly snapshot built from the measurements AI Speed Index has been collecting since launch. It does not claim to name a single permanent winner. It shows which fast AI models currently perform well enough, under documented conditions, to earn a place in the daily work of product managers, marketers, and small-team operators.
I treat the Index the same way I treated product scorecards in my earlier career: the summary is a convenience, the evidence underneath it is the substance. Readers who disagree with a weighting or a judgment should still be able to reach the raw tests and form their own view.
What the October Snapshot Contains
The first Index release includes every model that has completed the core test suite described in the scoring methodology article:
Baseline speed (first-token latency, tokens per second, total completion time, and consistency across ten runs)
Behavior under modest concurrent load
Cost calculated from observed token counts on real workflow tasks
Output quality notes on matched practical tasks
Reliability indicators (failure rate, truncation, recovery)
Results from at least one Real Workflows test
Models that still lack key measurements appear as provisional entries with the gaps clearly marked. No model receives a full composite score until the supporting data exists.
The October snapshot is dated. Pricing, model versions, and provider-side behavior can change. A later snapshot may reorder the same models once new measurements arrive. Historical snapshots remain accessible so that movement over time stays visible.
How to Read the Rankings
Each model entry shows five component scores and a weighted composite. The composite is useful for a quick scan. The components are more useful for decision-making.
A model that leads on speed but trails on cost may still be the right choice for interactive tools. A model that leads on cost and reliability but sits in the middle on pure latency may be the better fit for high-volume internal workflows. The Index is designed so those different priorities remain readable instead of being collapsed into one opaque number.
Every score links back to the underlying test articles. Readers can move from the summary table to the ten-run latency series, the cost-per-task calculations, the quality notes, and the workflow results without leaving the evidence trail.
What “Worth Your Time” Actually Means
The phrase is deliberately practical. A model is considered worth attention in the Index when it meets three conditions under the documented test conditions:
It delivers usable speed and consistency on the kinds of tasks the primary audience runs
Its observed cost stays inside a range that small teams can sustain
Its output quality and reliability reduce, rather than increase, the human effort required to finish the work
Models that post impressive peak numbers but show wide variance, frequent truncations, or high cleanup costs are scored accordingly. Models that look merely solid across the board yet remain predictable often rank higher for real use than models that shine in one dimension and stumble in others.
The Index does not claim that any model is universally best. It claims that the models currently listed have been measured on the same tasks, under the same conditions, and scored on the same transparent dimensions. That is enough to give readers a starting shortlist and a clear view of the trade-offs.
Caveats That Stay Attached to Every Snapshot
A monthly ranking still carries limits. The October Index does not claim:
That every fast model on the market has already been tested
That the tests capture every possible region, network, or load pattern
That the weights (Speed 25 %, Cost 20 %, Output quality 25 %, Reliability 20 %, Workflow fit 10 %) match every team’s priorities
That future model versions or pricing changes will preserve the current order
It claims only that the models included were measured with the published AI Speed Index methods, on the stated dates, and that the evidence remains attached to the scores. When those conditions change, the Index changes with them and the change is recorded.
How the Index Will Evolve
New models enter once they complete the core suite. Existing models are re-tested when material version or pricing updates appear. The monthly snapshot remains the public reference point; weekly data updates continue between snapshots as individual tests are finished.
Each new monthly release is accompanied by a short note that lists what moved, what was added, and what remains provisional. Readers who follow the Index over time will see both the current ranking and the trail of evidence that produced it.
The long-term goal is modest and useful: a living reference that small U.S. teams can consult when they need to decide which fast model is currently worth their time, their token budget, and their attention. The scores are a summary. The tests underneath them are the reason the summary can be trusted.
Fast is a feature. Reliable is the product. The October Fast Model Index is the first public expression of that distinction across the models AI Speed Index has measured so far. The list will grow, the scores will move, and the evidence will stay visible. That is the only ranking discipline worth maintaining.
Letters
No letters yet — be the first to write.