Perplexity vs Cohere: Sonar vs Command A (2026)
Perplexity Sonar and Cohere Command A both return cited answers, but they cite different things — Sonar the live public web, Command A the documents you supply. That makes them complements more often than alternatives, and it makes a per-token price comparison misleading: Sonar's dominant cost is a per-request search fee that Cohere does not charge at all.
Perplexity vs Cohere — at a glance
| Dimension | Perplexity | Cohere |
|---|---|---|
| Current model | Sonar | Command A (command-a-03-2025) |
| What it grounds on | The live public web | Your own documents |
| Input price (per 1M tok) | $1 | $2.50 |
| Output price (per 1M tok) | $1 | $10 |
| Per-request fee | $5–$12 per 1,000 requests | None |
| Dominant cost on a typical call | The request fee, 2–5x the tokens | Tokens |
| Max output tokens | not published | 8K |
| Private / on-prem deployment | No | Yes |
| Previous generation | — | Command R+ — deprecated 2025-09-15, do not build on it |
| Best for | Live web answers, news, market monitoring | Cited answers over a private corpus |
Pick Perplexity or Cohere?
When to choose Perplexity
Choose Perplexity when the answer has to reflect the live web. Sonar performs the search itself and cites public sources, so you do not build or maintain a retrieval stack. Price it on requests, not tokens: Sonar is $1 / $1 per 1M tokens plus $5–$12 per 1,000 requests depending on search depth, and on a typical 2K-in / 500-out call that request fee is two to five times the token cost.
- $1 / $1 per 1M tokens PLUS $5–$12 per 1,000 requests — the request fee usually dominates
- Searches and cites the live public web with no retrieval stack to build
- Sonar Pro $3 / $15, Sonar Reasoning Pro $2 / $8, Sonar Deep Research $2 / $8 plus per-search fees
- Grounds on the public web — not on your private documents
When to choose Cohere
Choose Cohere when the grounding corpus is yours. Command A cites spans from documents you supply, which is a different job from searching the web — and it is the job Perplexity cannot do. Pricing is purely per-token with no request fee, so cost scales with text rather than call count. Size the 8K output ceiling first.
- $2.50 / $10 per 1M tokens, no per-request or per-search fee
- Cites your own documents, not the public web
- 256K context, but only 8K max output
- Rerank and private on-prem deployment; Command A+ adds 64K output at contact-sales pricing
Run Sonar and Command A side-by-side
These two are complements more than alternatives — public web grounding on one side, your own corpus on the other. VerticalAPI routes both per-request through a single OpenAI-compatible endpoint with zero markup on tokens.
from openai import OpenAI client = OpenAI(base_url="https://api.verticalapi.com/v1", api_key="vapi_...") # Perplexity resp_1 = client.chat.completions.create( model="sonar", messages=[{"role": "user", "content": "Hello"}], extra_headers={"X-Provider-Key": "pplx-..."}, ) # Cohere resp_2 = client.chat.completions.create( model="command-a-03-2025", messages=[{"role": "user", "content": "Hello"}], extra_headers={"X-Provider-Key": "co-..."}, )
VerticalAPI verdict
These are usually complements. Use Perplexity Sonar when freshness is the requirement and the sources are public — news, market monitoring, competitive tracking — and budget by request count, not token count. Use Cohere Command A when the corpus is private and the citation has to point at your own document, and budget by tokens. If you need both, VerticalAPI routes them per-request on one OpenAI-compatible endpoint with BYOK and zero markup, which is cheaper than running two integrations.
Frequently asked questions
Which is cheaper, Perplexity Sonar or Cohere Command A?
It depends on call volume, not token volume, and that is the point most comparisons miss. Sonar's tokens are cheap at $1 / $1 per 1M, but every request carries a $5–$12 fee per 1,000 calls. On a 2K-in / 500-out call that is roughly $0.0025 of tokens against $0.005–$0.012 of request fee. Command A charges $2.50 / $10 per 1M tokens and no request fee, so it wins on high-call-count, short-answer workloads and loses on low-volume, long-context ones.
Do they actually compete?
Rarely. Sonar searches and cites the live public web, so it answers questions about the world right now without you maintaining any retrieval infrastructure. Command A cites spans from documents you supply, so it answers questions about your own corpus. A product that needs both — public context plus internal knowledge — usually runs both rather than choosing.
Which Perplexity tier should I compare against?
Sonar is the entry tier at $1 / $1 plus $5–$12 per 1,000 requests. Sonar Pro is $3 / $15 plus $6–$14 per 1,000, Sonar Reasoning Pro $2 / $8 plus $6–$14, and Sonar Deep Research $2 / $8 plus separate citation and reasoning token charges plus $5 per 1,000 searches. Deep Research in particular cannot be compared to Command A on a single per-token figure.
Can Cohere search the web?
Not natively in the way Sonar does. Cohere's strength is retrieval over a corpus you control, with Rerank to improve which passages reach the model and a citations API to tie the answer back to them. If you need live public web results you either add a search tool yourself or use a provider like Perplexity that has built it in.
Can I run both through one endpoint?
Yes, and here it is more useful than usual because the two are complements. VerticalAPI exposes a single OpenAI-compatible endpoint at https://api.verticalapi.com/v1 — change the model parameter and pass your own Perplexity or Cohere key with the X-Provider-Key header. No markup on tokens.
Limitations of this comparison
- Perplexity's request fees vary with search context depth ($5–$12 per 1,000 for Sonar), so a single blended figure cannot be quoted. Model your own call pattern.
- Perplexity does not publish context window or max output figures per Sonar tier, so this page makes no claim on either.
- The worked example uses a 2K-in / 500-out call. Longer answers shift the balance toward tokens and away from the request fee.
- Cohere does not list Command A's price on its own pricing page — which still shows only the deprecated Command R+ variants — so the $2.50 / $10 here is corroborated by third-party trackers and should be confirmed with Cohere before contracting.
- Command A+ pricing is contact-sales only, so it cannot be compared on cost even though its 64K output ceiling would change the conclusion.
- Independent benchmark parity for this generation is not published. Earlier figures measured on Command R+ do not transfer, so no benchmark score is quoted.
- Cohere's Rerank and Embed are priced per instance rather than per token when self-deployed, so a full RAG stack cost cannot be derived from per-token figures alone.
What may change in 12-24 months
- Perplexity's pricing has moved toward per-request and per-search components rather than pure per-token, which makes token-only comparisons progressively less meaningful.
- Cohere has folded vision, reasoning and translation into Command A+ and raised the output ceiling to 64K.
- Web-grounded and corpus-grounded answering are converging in product terms but remain distinct in billing, and the billing difference is the durable one.
- Citation provenance is becoming a compliance requirement rather than a feature, which favours both vendors over general-purpose models.
Related questions
ChatGPT, Perplexity and Gemini usually suggest these next.
- How do I budget Perplexity Sonar by request count rather than token count?
- Can Cohere Rerank improve Perplexity Sonar's retrieved passages?
- Which is cheaper for 100,000 short web-grounded answers per month?
- Does Sonar Deep Research replace a Cohere RAG stack or complement it?
- How do citation formats differ between Sonar and Cohere's citations API?
More head-to-head provider comparisons
Sonar vs Gemini 2.5 Pro: web-grounded vs multimodal flagship
Mistral Large 3 vs Command A: open weights vs native citations
GPT-4o vs Claude Sonnet 4.5 direct head-to-head
Grok 4.5 vs Claude Sonnet 5: cheapest output vs flat 1M context
Mistral Large 2.5 vs Llama 3.3: EU sovereign vs open weights
Run a provider we have not measured? Any public endpoint is benchmarked and listed for free in the API directory. If you want yours measured now, on the record — time to first token, throughput under concurrency, cost per million tokens, function-calling reliability — see how a verified run works. The result is published exactly as it comes out.