Two of the biggest names in AI shipped the same idea on 6 October 2026: a model that does not write anything. OpenAI launched a Decisions API that returns typed answers, such as the probability that something is true, a pick from a fixed list or a score against a rubric [1][2]. Perplexity released pplx-decider-v1.1-27b, an open-weight model that does the same job [5][6]. OpenAI also cut its API usage tiers from five to three [1][3].
All of this is confirmed by the companies. Speed and benchmark claims are the vendors’ own, and we label them that way.
What is a decision model?
Most AI products are generative: they write an answer word by word. A lot of real work does not need prose. A support desk needs to know which team should get a complaint; a shop needs to know whether a product photo shows damage.
A decision model answers such questions directly. You supply the evidence and a typed question; it returns numbers. OpenAI offers three question types [2]:
- Predicate: the probability, from 0 to 1, that a condition is true.
- Choice: one option from a list you supply, with a probability for each.
- Score: a position on ordered levels you define, such as “cosmetic”, “workaround available” and “fully blocked”.
Perplexity’s API offers the same three shapes and calls its yes/no type “noul” [8], the name TypeSafe AI uses for its own decision model, Jev, which TSN explained last week.
The model estimates; your code decides. OpenAI tells developers to set thresholds from labelled examples in their own application, weighing the cost of false positives and false negatives [2].
What has OpenAI launched?
The Decisions API is in public beta, with general availability expected “in the coming weeks” [2]. It runs on one model, gpt-6-luna, the smallest of the GPT-6 family, and accepts text, images or both [2].
The price is $0.10 per million input tokens, and nothing else: no output-token, cache-read or cache-write charges, though regional processing premiums and long-context multipliers still apply [2]. (A token is a small chunk of text.) Used normally, the same Luna model costs $0.10 per million input tokens and $0.50 per million output tokens [4].
OpenAI says the API answers about 10 times faster than its Responses API [1][2]. That is OpenAI’s claim; the material we read did not say how it was measured. If you need generated output, such as extracted fields or a written explanation, OpenAI points you to its existing Structured Outputs feature instead [2].
What changed in OpenAI’s API tiers?
OpenAI has “simplified API usage tiers from five to three”: Build, Launch and Grow [1]. Organisations move up automatically as total credit purchases pass each threshold [3]:
- Build: $5 in total credit purchases; $500 a month usage limit.
- Launch: $100; $5,000 a month.
- Grow: $500; $200,000 a month.
A Free tier for users in eligible countries remains, with a $100 monthly limit, and higher tiers bring higher rate limits [3].
What has Perplexity released?
pplx-decider-v1.1-27b is a 27-billion-parameter model built on Alibaba’s Qwen3.8-27B and published on Hugging Face under the permissive Apache 2.0 licence [5]. Instead of a text-writing output layer, it has a separate “decision head” that turns its reading of the input into probabilities over the valid answers [5]. Perplexity says most of the gain over v1 came from letting the model read the whole input at once rather than strictly left to right (“lifting of the causal mask”) and from more training data [5]. Running it yourself needs a GPU with room for roughly 49 GiB of weights [5].
Perplexity’s own figures put v1.1 at 61.56 on the Decision Index, up from 56.4 for v1 and ahead of Jev’s 57.9 [5]. The Decision Index is a community leaderboard on Hugging Face that tracks open attempts to reproduce Jev [6][7]. The leaderboard’s own latest data, generated on 7 October under a newer scoring version (v0.3), shows different numbers: 62.75 for Perplexity’s model and 60.11 for Jev, still with Perplexity first [7]. The formula changed between editions, so the two sets are not directly comparable, and we have not checked how either was produced.
Through Perplexity’s API the model costs $0.02 per million input tokens, with free output and no per-request fee [8]. Perplexity’s announcement calls that half the price of v1 [6], but its pricing page, as we read it on 8 October, lists both versions at $0.02, so we could not confirm v1’s earlier price [8].
Why it matters
Decision models skip the step where a model writes text that a program must then read back: no prose to parse and, on these price lists, no output to pay for. OpenAI offers a hosted service on a closed model; Perplexity offers a cheaper hosted service plus weights companies can, in principle, run on their own hardware. We have seen no head-to-head test, and the price gap says nothing about which answers better for a given job.
What this does not prove
- That “10x faster” holds for your workload. It is OpenAI’s comparison with its own Responses API [1][2].
- The Decision Index scores. The card figures are Perplexity’s own, and the leaderboard’s latest numbers differ [5][7].
- That the weights are freely downloadable today. Hugging Face lists the repository as public and Apache 2.0, but the card’s usage notes mention a “private repository” and the leaderboard notes “repo access by request” [5][7]. We did not download it.
- Who gains from the new tiers. We did not have the old five-tier thresholds to compare.
- That a probability is a correct answer. A confident wrong pick is still wrong, which is why thresholds are left to developers [2].
The Bottom Line
OpenAI and Perplexity have both argued, on the same day, that many AI tasks need a number, not an essay. OpenAI’s Decisions API runs on GPT-6 Luna at $0.10 per million input tokens with no output charge, alongside simpler Build, Launch and Grow tiers [1][2][3]. Perplexity’s Apache 2.0 decider costs $0.02 per million input tokens and leads a community leaderboard on Perplexity’s own figures [5][6][8]. The idea is sound; independent tests on real workloads will show how well it works.
Related on TSN: RAG vs Jev+RAG Explained: What the Judgment Gate Adds · Small Open Models That Answer in One Pass: Liquid AI’s d1 and Perplexity’s pplx-embed-v2-late (link to be added when live).
Sources
- OpenAI, “Changelog,” OpenAI API documentation (entries dated 6 October 2026: Decisions API beta; usage tiers simplified from five to three). https://developers.openai.com/api/docs/changelog
- OpenAI, “Decisions,” OpenAI API documentation (guide; question types, beta status, pricing and availability), read 8 October 2026. https://developers.openai.com/api/docs/guides/decisions
- OpenAI, “Rate limits” (usage tiers table), OpenAI API documentation, read 8 October 2026. https://developers.openai.com/api/docs/guides/rate-limits
- OpenAI, “Pricing,” OpenAI API documentation (gpt-6-luna standard rates), read 8 October 2026. https://developers.openai.com/api/docs/pricing
- Perplexity, “pplx-decider-v1.1-27b” model card and repository metadata (licence apache-2.0; base model Qwen/Qwen3.8-27B), Hugging Face, read 8 October 2026. https://huggingface.co/perplexity-ai/pplx-decider-v1.1-27b
- Perplexity (@pplx-announcements), “Pplx-decider-v1.1-27b, our updated open weights multimodal decision model, is now available,” Perplexity Community, 6 October 2026, 22:11 BST. https://community.perplexity.ai/t/pplx-decider-v1-1-27b-our-updated-open-weights-multimodal-decision-model-is-now-available/6286
- multimodalart, “Jev Decision Index” (community leaderboard Space; data file generated 7 October 2026, 19:25 BST), Hugging Face. https://huggingface.co/spaces/multimodalart/jev-decision-index
- Perplexity, “Pricing” (Decisions API table) and “Answer questions” (Decisions API reference), Perplexity API documentation, read 8 October 2026. https://docs.perplexity.ai/docs/getting-started/pricing · https://docs.perplexity.ai/api-reference/decisions-post
