DeepSeek V4 Pro vs Grok 4.6: A Price War Fought 24 Hours Apart, and Who Actually Won It

Grok 4.6 shipped on August 12, 2026, and DeepSeek V4 Pro followed the next day — the two flagship launches of the same forty-eight hours, now a month behind us and newly worth revisiting because one of them spent last week trying to retire and then changing its mind. Both are frontier-class reasoning models, but they could not be more different products: Grok 4.6 is x​AI's proprietary flagship at $2.00 per million input tokens and $6.00 per million output tokens, and DeepSeek V4 Pro is the MIT-licensed open-weights model from Deep​Seek at $0.66/$1.98 off-peak (or $1.32/$3.96 at peak). The interesting thing about the matchup is how little the raw price ratio — about 3x on output, over 9x when you stack V4 Pro's off-peak and cache windows — tells you about who should use which. The two are aimed at different buyers, and the September events on the Deep​Seek side changed the calculus for exactly one of those buyers.

Independent numbers in this piece are Artificial Analysis measurements. The vendor-run figures are labeled as vendor-reported because neither company's launch benchmarks have been reproduced by a third party.

Same week, same reasoning class, opposite philosophies

On the Artificial Analysis Intelligence Index, Grok 4.6 scores 44 at high effort — ranked #20 of 200 across the full index — while DeepSeek V4 Pro scores 36, ranked #7 in its own class. An eight-point gap, which means the premium does buy measurable reasoning, not just packaging. But it is the packaging that explains the rest of the price difference: Grok 4.6 is a multimodal proprietary model — text and image input, 500K context, reasoning by default — that you call through x​AI's API or a partner. DeepSeek V4 Pro is text-only, 1M context, and shipped with MIT-licensed weights you can download and serve yourself.

• Intelligence Index — Grok 4.6 44 (#20/200) vs V4 Pro 36 (#7/113)

• Price — Grok 4.6 $2.00/$6.00 per 1M vs V4 Pro $0.66/$1.98 off-peak ($1.32/$3.96 peak)

• Speed — Grok 4.6 ~57 tokens/sec vs V4 Pro ~81 tokens/sec

• Context — Grok 4.6 500K vs V4 Pro 1M

• Input — Grok 4.6 text and image vs V4 Pro text-only

• Weights — Grok 4.6 proprietary vs V4 Pro MIT open weights

Where the 3x price premium does and does not show up

The price gap is real and it is partly an intelligence gap, but only partly. Grok 4.6 leads by eight points on the independent index, and the rest of the premium buys you things other than raw ability: x​AI's ecosystem, image input, and a 500K context served by a company whose whole business is that one product. The stronger contrast is the deployment story. Grok 4.6 is a service: you rent it, you never own it. DeepSeek V4 Pro is also a service, but the MIT weights mean it is also infrastructure: if the API price or the vendor's roadmap stops suiting you, you can stand up the same model on your own hardware for the marginal cost of compute. That is a difference in kind, not degree, and it is the one that makes the "cheaper" framing incomplete.

The September reversal changed one half of this story

The week of September 8, Deep​Seek put V4 Pro on a retirement path: from September 14, requests to deepseek-v4-pro would be routed to V4.1 Flash and billed at Flash prices until a V4.1 Pro shipped. Then it delayed the cutoff, then on September 11 it reversed entirely, committing to keep serving V4 Pro at unchanged billing and saying it would give notice if that ever changes. The reversal was confirmed by Chinese state media and multiple tech outlets the same day. For this matchup, that news matters on the Grok side more than the Deep​Seek side: anyone who had been leaning Grok 4.6 because "the cheap open model is about to vanish" had their justification quietly deleted. The affordable half is durable, which makes the premium half a choice rather than a forced hedge.

Who the matchup actually favors

If you want a frontier reasoning model with image input, a long context, and no interest in running your own weights, Grok 4.6's $2/$6 is fair value and its 44 on the index is competitive with anything at that tier. If your workload is text-only and token-heavy, DeepSeek V4 Pro is the better answer on nearly every axis that shows up in the bill: roughly three times cheaper at peak, deeper cache discount, faster output, twice the context, and open weights as an escape hatch. The genuinely hard case is text-only and reasoning-heavy with tight latency needs — there the eight-point gap matters, and your eval should decide, not the marketing.

Both are on OrcaRouter today — Grok 4.6 and DeepSeek V4 Pro both route through the platform — so you can call either (or both) on one key, with provider list prices passed through at zero markup. Because the pass-through is exact, the price columns in this article are the prices you actually pay, and the routing DSL means you can send image-input tasks to Grok 4.6 and long-text tasks to DeepSeek V4 Pro out of the same integration. The piece's working assumption is that you would want both precisely because neither matches the other's strengths.

The verdict

Grok 4.6 and DeepSeek V4 Pro are both strong reasoning models with genuinely comparable independent scores, and the price gap between them buys a real-but-narrow edge in intelligence plus an entirely different product philosophy: multimodal input and a closed ecosystem versus text-only, open weights, and a three-to-nine-times cheaper token. If you are renting reasoning, either will do and your eval decides. If you are building infrastructure, DeepSeek V4 Pro is the model that a week of retirement drama could not kill — and as of September 15, it is also the one with a written promise to stay.