Perplexity Decider v1.1 Cuts Decisions API Price to $0.02

A second release in under a week

Perplexity has shipped an update to its open-weight decision model less than a week after the original launch. On October 6, 2026, the company announced pplx-decider-v1.1-27b, which it describes as an updated multimodal decision model with open weights. Input pricing drops to $0.02 per million tokens, half the cost of v1. 1 The company also says v1.1 posts the top score on the new Hugging Face Decision Index 0.3 benchmark. 1

The original model, pplx-decider-v1-27b, launched on October 1 alongside a new Decisions API. 24 Coverage at the time quoted CEO Aravind Srinivas, who said Perplexity intended "to bring down the price even further over the coming days." 3 The v1.1 release follows through on that promise almost immediately.

What the Decisions API actually does

The product is unusual. The Decisions API does not generate free-form text. It returns a probability distribution over a fixed set of answers the developer defines in advance. 3 Responses come in three typed shapes: yes/no, multiple choice, and ordered score. 4 A request can also score something against a rubric. 3

The practical pitch is reliability in automated pipelines. A general chat model asked to sort a ticket into "billing, technical or sales" might reply with a hedged sentence that breaks the code parsing it. The decider can only return one of the permitted options. 3

On the technical side, the model is a fine-tune of Qwen3.8-27B. It is released on Hugging Face under Apache 2.0 and has a 262,144-token context window. 4 According to Perplexity's documentation, latency stays under two seconds for inputs of a few hundred tokens and grows to about 23 seconds near the maximum input length. 4 Output tokens are free. The original input price was $0.04 per million. 34

The benchmark picture

At launch, Perplexity published results on its own 11-benchmark panel of 7,210 rows, dated September 2026. 2 The scores were:

• pplx-decider: 85.71%
• Jev 1.13.0: 84.51%
• Qwen3.8-27B (the base model): 74.76% 24

The wins were not uniform. The largest margin came on RAGTruth, a hallucination-detection test, where the decider scored 88.80% against Jev's 77.27%. 4 It also led on FinancialPhraseBank, TabFact and Circa. 2

Jev came out ahead on several reasoning-heavy tests:

• BBH: 94.27% versus 82.80%
• WinoGrande: 90.70% versus 83.30% 2

On JevBench's public hard set, the decider trailed both Jev and its own Qwen base model. 2 The overall result is a narrow aggregate lead built mostly on classification and grounding tasks, not general reasoning.

The v1.1 claim uses a different yardstick. Decision Index 0.3 ranks Perplexity Decider v1.1 first at 62.8, with Jev 1.13.0 at 60.1 as a reference point. 2 One reporting outlet dates the index's opening to October 7, while Perplexity's announcement referencing it is timestamped October 6. 12 The one-day gap likely reflects time zones or a staged rollout, but it shows how closely the release was coordinated with the benchmark's debut.

Reading the move

Three points stand out.

The panel was Perplexity's own. The 85.71% headline number came from a panel the company assembled, and the lead over Jev was barely more than a point. 24 Decision Index 0.3 is a third-party-hosted benchmark, so it is arguably a more meaningful signal. Still, a 2.7-point lead on a brand-new index is a thin basis for strong conclusions about real-world performance.

Price may matter more than the scores. At $0.02 per million input tokens with free output, cost is close to a rounding error for most classification workloads. 12 That changes the comparison with a general-purpose LLM. Developers do not have to decide whether the decider beats a frontier model on reasoning. They can ask whether it is good enough for routing, moderation or grounding checks at a fraction of the cost.

The open weights reinforce the price play. Because the model is Apache 2.0 licensed, anyone who would rather self-host can do so. 34 Perplexity seems to be betting that the hosted API wins on convenience and that open weights build trust in the approach.

Taken together, this looks less like a model race and more like an attempt to make "structured decision" a standard, cheap API category before competitors define it. Shipping v1.1 within days, at half the price and with a fresh benchmark win, signals that Perplexity wants that position quickly. Whether the decider holds up outside curated panels is something independent developer testing will have to show.