August 21–22, 2026, OpenAI cut official GPT-5.6 Sol API and credit pricing by more than 20% — $4 / $20 per million input/output tokens through at least November 21. Aggregators reframed the story as "Perplexity cuts Sol Computer pricing 20%" because Perplexity Computer and the Gateway API route Sol at upstream published rates, not a bespoke Perplexity subsidy.
If you run agentic loops in Perplexity Computer, your Sol-shaped bill just moved — but only because OpenAI moved the price card, not because Perplexity self-funded a discount like the earlier OpenRouter non-BYOK promo.
TL;DR — what people are asking
| Question | Direct answer |
|---|---|
| Who cut the price? | OpenAI — official API/credit table, Aug 21–22 |
| New Sol rate? | $4 in / $20 out per 1M tokens (was $5 / $30) |
| Perplexity's role? | Pass-through on Gateway; Computer inherits upstream |
| Same as OpenRouter 50% badge? | No — that was router promo; this is OpenAI official |
| Subscription tiers? | Unchanged for ChatGPT Pro/Plus/Business; Perplexity flat fees separate |
| Promo end? | ≥ Nov 21, 2026 per OpenAI |
| vs Fable API? | Sol now undercuts Fable list ($10/$50) on paper |
The billing chain for Perplexity Computer
Perplexity Computer is a productized agent harness — Search as Code, Brain memory, connectors, browser automation — not a single model weights file. When you pick GPT-5.6 Sol in Computer or call openai/gpt-5.6-sol through Gateway, you are consuming OpenAI-metered tokens plus Perplexity's orchestration overhead.
Perplexity's Gateway positioning (also covered when Grok 4.6 joined the Agent API) is transparent per-token pricing at provider-published rates — no extra per-request fee. That design means:
- OpenAI cuts Sol → your Sol line item drops without Perplexity issuing a parallel press release.
- OpenAI raises DeepSeek competitor prices elsewhere → Gateway reflects that too.
- Orchestrator swaps (GLM 5.2, Grok 4.6) remain Perplexity's lever when upstream Sol is still too expensive for a task class — see GLM orchestrator at 0.34× Opus.
Why output pricing matters more for Computer
OpenAI's cut is asymmetric: ~20% on input, ~33% on output. Agent harnesses — Computer included — burn output tokens on tool JSON, planning prose, and multi-step retries. That is the same structural reason loop engineering guides push cheap planners + expensive escalations.
Rough intuition for a 10:1 output:input agent task at old vs new Sol list:
| Old ($5 / $30) | New ($4 / $20) | |
|---|---|---|
| 100k in + 1M out | $0.50 + $30 = $30.50 | $0.40 + $20 = $20.40 |
| Savings | — | ~33% on this skew |
Your mileage varies with Search-as-Code retrieval and connector calls — measure on your task mix.
Comparison anchors on explainx.ai
- OpenAI official Sol cut (Aug 22) — primary source breakdown
- OpenRouter 50% promo ≠ OpenAI cut — avoid double-counting discounts
- Fable vs Grok vs Sol vs Qwen shootout — quality per dollar after price moves
- DeepSeek V4 vs GPT-5.6 live comparison — open-weight pressure on the same tier
Honest limitations
- Perplexity has not, as of this writing, published a dedicated blog post isolating "Computer Sol −20%" — the story is inferred from Gateway economics, which is correct but worth stating.
- Max/Pro inclusion of Sol vs orchestrator defaults can change product-side without API list moves — check Perplexity's in-app model picker.
- Rate limits complaints followed OpenAI's cut on direct API — Perplexity may hit different capacity pools.
- Enterprise agreements may not track public list automatically.
What this means for what you build or pay
Teams already on Perplexity Max + Computer for research agents should re-benchmark Sol against GLM/Grok orchestrators — Sol may now win tasks you previously routed away for cost. Teams on pure API should compare direct OpenAI vs Gateway on failover, logging, and web-search grounding benchmarks, not just list price — pass-through only helps if the router path is what you need.
Related on explainx.ai
- Perplexity portable Computer on DGX Spark — local agents demo (Aug 26)
- OpenAI GPT-5.6 Sol official API price cut
- Perplexity Computer GLM 5.2 orchestrator cost math
- Perplexity Computer Projects — multiplayer agentic OS
- Perplexity adds Grok 4.6 — verified pricing
- AI token pricing explained
- Att model routing — 56% coding cost reduction pattern
- OpenRouter enterprise guide
- alphaXiv's orx-figures skill for GPT-5.6 Sol / Codex research figures
Sources: OpenAI Developer Community announcement (August 21, 2026), Reuters, OpenAI pricing table, explainx.ai prior OpenRouter promo coverage. Rates and promo end date accurate as of August 25, 2026 — verify live model pages before budgeting production Computer workloads.