DeepSeek V4 Flash vs Gemini, GPT and Claude: Similar AI Scores, Vastly Different Token Prices

On Artificial Analysis’s cross-vendor index, DeepSeek V4 Flash and Gemini 3.6 Flash both score 52. Against DeepSeek’s off-peak rates, Google charges 3.4 times more for input and 5.7 times more for output.

Those are list prices per token. What a given task costs depends on how many tokens each model spends reaching an answer, which no price table shows — so read 5.7x as the gap on the output rate, not a promise about your bill.

That is the comparison worth making, and it is not the one you get by matching tier names. Matched on measured capability instead, DeepSeek’s peers are Gemini 3.6 Flash, GPT-5.6 Luna and Claude Sonnet 5 — not the models with “Flash” or “Haiku” in the name.

What does DeepSeek V4 Flash actually cost?

Six numbers now, from DeepSeek’s own pricing page, because its rates are published in two bands. All are per 1M tokens.

The model version is DeepSeek-V4-Flash-0731. It carries a 1M-token context window and a maximum output of 384K tokens, supports both thinking and non-thinking modes with thinking as the default, and has a concurrency limit of 2,500.

For reference, the tier above it — DeepSeek V4 Pro, now at version DeepSeek-V4-Pro-0813 — costs $0.66 input, $1.98 output and $0.022 on a cache hit off-peak, with a concurrency limit of 500.

Those figures have a history worth knowing, because it shows how movable these numbers are. On 1 May 2026 the same page listed Pro at a 75% discount: $0.435 against a standard $1.74, and $0.87 against $3.48. The discount markers went, and the rates have moved again since — Pro is now $0.66 and $1.98 off-peak. V4 Flash has moved too. It was a flat $0.14 and $0.28 when this article was first written on 31 July, with no discount marker on it at all; it is $0.22 and $0.66 off-peak today.

Which models is it actually competing with?

Not the ones with matching names. That is the easy comparison and it is the wrong one.

Matching by tier label — DeepSeek’s “Flash” against everyone else’s “Flash”, “Luna” or “Haiku” — assumes vendors use those words to mean the same thing. They do not. To place these models on one scale we used the Artificial Analysis Intelligence Index , which runs its own evaluations across vendors on a single methodology, rather than each vendor’s self-reported benchmarks, which are not comparable with each other.

On that index, DeepSeek V4 Flash scores 52. Here is where it sits, with Gemini 3.5 Flash-Lite kept in the bottom row for contrast rather than as a peer:

Two corrections fall out of that, and both were in an earlier version of this article.

Gemini 3.5 Flash-Lite is not a peer. It scores 37 against V4 Flash’s 52 — fifteen points below, five times the three-point gap that separates V4 Flash from Claude Sonnet 5 at the top of these tables. Including it because its price was close was exactly the error of matching on the wrong axis.

Claude Haiku 4.5 is not a peer either. The index does carry a score for it, on the model’s own page — an earlier version of this article wrongly said it did not — and when we read that page in July, the score sat far below every model in this comparison. It has not been re-read under the new index version, so no figure for it is quoted here: it is a real model with a real score, and that score is nowhere near this comparison.

Anthropic is the awkward case, and worth setting out honestly. The two models in these tables sit eighteen points apart, and neither lands just below DeepSeek V4 Flash:

Sonnet 5 is the nearer model on both axes at once: three points above V4 Flash, and cheaper than Sonnet 4.6 at $2 and $10 against $3 and $15. Sonnet 4.6 is fifteen points below V4 Flash and half again the price of the better model above it, which leaves it out of the comparison on both counts rather than only one. Sonnet 5’s $2 and $10 were announced as introductory rates expiring on 31 August 2026; Anthropic has since made them the standard price. So the tables below use Sonnet 5, which is what anyone choosing today would actually pick — and nothing in Anthropic’s range now sits just under DeepSeek V4 Flash. On the index version this article was first written against, Sonnet 4.6 did; the re-versioned index moved it a long way down.

At matched capability, what does each one cost?

As multiples of DeepSeek V4 Flash’s off-peak rates:

The Gemini row is the one to sit with. Gemini 3.6 Flash scores 52, exactly level with DeepSeek V4 Flash, and costs 3.4 times as much on input and 5.7 times as much on output off-peak. Level on one index is not level on everything — but there is nothing in these two scores to pay 5.7x for, and Google’s own page has that gap widening on 1 January 2027, to 6.8x and 11.4x.

GPT-5.6 Luna is the genuinely competitive answer, and it has moved past DeepSeek on one side: level on the index at 52, at 0.9x input — cheaper, in other words — and 1.8x output. Claude Sonnet 5 buys three index points at 9.1x and 15.2x, and those are now its permanent rates rather than introductory ones.

What the step up to V4 Pro buys

DeepSeek V4 Pro costs three times V4 Flash and scores one point above it on this index: 53 against 52, at $0.66 input and $1.98 output against $0.22 and $0.66, off-peak on both.

That is one index point for exactly three times the price, in both bands. And one point is as small a difference as this index can now express — v4.1.1 publishes whole numbers, so what sits behind that point may be narrower than it looks. Pro may well win on things the index does not capture — but on the index, a single point is all it wins, and it is not obviously worth 3x. Earlier versions of this article had Pro scoring below Flash on the old index and called it the worse buy outright. On the current reading that reasoning was wrong — Pro is ahead, barely — and the case for staying on Flash now rests on the price rather than on the score.

What the capability match does not settle

Three things, and they matter enough to state before anyone acts on the tables.

Reasoning effort is not held constant. DeepSeek V4 Flash, GPT-5.6 Luna and Claude Sonnet 5 are all measured at max; Gemini 3.6 Flash is measured at high, as its own model page states. Different settings, not a missing one. Scores taken at different effort settings are not strictly like for like — and effort drives output tokens, which is the expensive side of every price above. A model that reaches its score by thinking longer pays for it twice.

This is one index. Artificial Analysis is a single evaluator with a single methodology. It is the right kind of source for this question — the same harness across vendors, rather than four sets of self-reported numbers — but a different evaluator would order these models differently, and none of them measures your workload.

Index points are not tokens. A model that scores a point higher but needs half as many attempts to get a task right is cheaper in practice than the per-token table suggests. Cost per finished task is the number that decides a bill. Artificial Analysis does publish a version of it — a cost per Intelligence Index task, and output tokens per task, which fold reasoning tokens and cache behaviour into the total — but the figures sit in charts we could not extract reliably, so none are quoted here. That metric, not the rate card, is where this question is actually settled.

Three dated changes, and where they landed

All three are published by the vendors themselves. Two of them have already happened, and in opposite directions.

DeepSeek’s peak pricing is in force. In July the footnote on its pricing page said a peak/off-peak policy was coming at 2x the regular rates, with the effective date still “subject to the official announcement”. It has since arrived. The page now publishes every rate twice, and gives peak hours as 01:00–04:00 and 06:00–10:00 UTC, Monday through Friday, with all other hours off-peak — the same seven hours a day the footnote described, now on weekdays only.

The consequence is worth spelling out. At peak, V4 Flash costs $0.44 input and $1.32 output, which puts it above GPT-5.6 Luna on both sides. Off-peak it is well under Luna on output, $0.66 against $1.20, but its $0.22 input is already above Luna’s $0.20. The “cheaper on both sides” line this article ran in July does not survive either band.

Nor is DeepSeek the only one moving. The $0.20 and $1.20 used here are the figures published on OpenAI’s pricing page on 26 August 2026; that page carries no change log, so a comparison against an older capture of it may not reconcile, and the date this article was read is the only anchor we can offer.

Claude Sonnet 5’s increase was cancelled. Anthropic’s pricing page used to carry two rows for the same model: $2 input and $10 output per MTok through 31 August 2026, and $3 and $15 from 1 September. The second row has gone, replaced by a note saying the scheduled increase “will not occur” and that the $2/$10 launch pricing is now the standard price. Anthropic’s output multiple therefore stays at 15.2x rather than rising by half, and the Sonnet 5 figures in the tables above are permanent rather than introductory.

Google’s is the dated change still ahead. Gemini 3.6 Flash’s published rates run to a deadline: $0.75 input and $3.75 output through 31 December 2026, then $1.50 and $7.50 from 1 January 2027. Cached context doubles with them, $0.075 to $0.15, and so does the storage fee, $0.50 to $1.00 per million tokens per hour. It is the one dated row still inside the comparison table.

One other thing on Anthropic’s page that surprises people: Claude Fable 5 is among its most expensive models at $10 input and $50 output, double Opus 5’s $5 and $25. It is not alone at that price — Claude Mythos 5, listed as limited availability, matches it — and the retired Opus 4.1 and Opus 4 are still listed higher at $15 and $75.

What these numbers do not tell you

Four things, in rough order of how much they move a real bill.

Cache mechanics are not comparable. DeepSeek prices a cache hit at $0.007 off-peak and $0.014 at peak, and applies it automatically. Anthropic charges separately for writing to cache — on Sonnet 5, $2.50 per MTok for a 5-minute cache and $4 for an hour — before you get the $0.20 hit rate, and all three are now permanent rather than introductory. Google charges $0.075 for cached context plus a storage fee of $0.50 per million tokens per hour, both doubling on 1 January 2027. Three different billing models wearing the same word. The single “cheapest cached input” column above is the only piece of them that can be lined up.

Context tiers. OpenAI charges a different rate once a request crosses into its long-context band — GPT-5.6 Luna goes from $0.20/$1.20 to $0.40/$1.80. DeepSeek publishes one rate for its full 1M-token window, whichever band you are in. For long-document work that difference compounds quietly.

The free tier. Google publishes a free tier for Gemini alongside the paid one, which no amount of per-token comparison captures.

List price is not the bill. Reasoning tokens you never see, cache-hit rates in the real world, retries, and per-request overheads all sit between the rate card and the invoice. We took that apart separately in what Opus 5, Fable 5 and GPT-5.6 Sol actually cost , and the reasoning there applies to every model here.

And one thing this article deliberately does not do: rank these models on quality. Price per token says nothing about how many tokens a model needs to get an answer right, which is the number that actually decides cost per finished task.

So is DeepSeek V4 Flash the cheapest?

On output and on cached input, off-peak, yes — by a wide margin. On plain input, no: GPT-5.6 Luna’s $0.20 now undercuts DeepSeek’s $0.22.

Off-peak and on output, nothing here is close to it: Luna, the nearest, is 1.8x. On input Luna is now the cheaper of the two in both of DeepSeek’s bands. On cached input its peers charge 2.9x to 28.6x more off-peak, and 1.4x to 14.3x against DeepSeek’s peak rate. That matters most for agent-style workloads that re-send the same context repeatedly.

The honest summary is narrower than the headline. DeepSeek V4 Flash is priced like a tier below the models it actually scores alongside, and its advantage is widest on output tokens and cache hits — which is where reasoning-heavy and context-resending workloads tend to spend, though not where every workload spends.

Sources

Each vendor’s own pricing page, all read on 26 August 2026.

Anthropic’s pricing documentation is reached through redirects from docs.anthropic.com and docs.claude.com; the final URL is the one cited. Vendor pricing pages change without notice — every current price here is as published on 26 August 2026, and the two earlier sets of DeepSeek rates are cited to the dated archive captures above. The Intelligence Index scores carry the same date: they were re-read on 26 August 2026, under index version v4.1.1, and replace the 31 July figures this article previously carried. Every table carries the date its own figures were read.