Output tokens per deck:
3 results · tokens, lower is better. Choose a model to highlight it.Clear highlight
- 1 DeepSeek-V4.1-Flash-1017DeepSeek 21,019
- 2 Mercury 3Inception 32,448
- 3 MiniMax M3.1 FlashMiniMax 45,141
Benchmarks / DeckBench
Measured by Spring Prompt
DeckBench
Given a finished analysis, which models make a deck a manager could present as it is: true to the numbers, telling the right story, and well designed?
Which AI model makes the best business presentation? Compare the results below, then inspect and download the slide images, read the judges' findings and compare models' decks. DeckBench tests underlying models on the same business briefs; presentation apps add their own templates and editing tools.
Last updated 19 Oct 2026
- Results dated
- 19 Oct 2026
- Models
- 3
- Unit
- tokens
- Licence
- Spring Prompt original
- Judge
- GPT-6.1 Sol, Claude Opus 5.5 and Gemini 3.1 Pro, by majority (jp-0f0755d8be1b, jp-29cf93e00843) · bias check
- Runs
- 1 deck per task, 6 tasks
Deck rating against cost
What one deck cost at the run date's prices, on a log scale. Up and to the left is better.
Named: the six best and the best for the moneyOther models (hover for names)Best score at each cost
Best per budget
- GPT-6 Luna: 1,343 at $0.0087
- Claude Sonnet 5.5: 1,408 at $0.15
- GPT-6.1 Sol: 1,457 at $0.15
- GPT-6 Sol: 1,528 at $0.21
- GPT-6 Astra: 1,577 at $1.09
Cheapest first: each model here beats every cheaper one on score.
Full results
| # | Model | Output tokens per deck tokens, lower is better | Deck rating rating | Presentable decks % of tasks | Accurate decks % of tasks | Design quality % of the maximum | Cost per deck US dollars |
|---|---|---|---|---|---|---|---|
| 1 | DeepSeek-V4.1-Flash-1017DeepSeek |
21,019
|
1,145 | 0.0% 0 of 6 | 33.3% 2 of 6 | 68.5% | $0.0255 |
| 2 | Mercury 3Inception |
32,448
|
812 | 0.0% 0 of 6 | 0.0% 0 of 6 | 58.5% | $0.0116 |
| 3 | MiniMax M3.1 FlashMiniMax |
45,141
|
1,090 | 0.0% 0 of 6 | 50.0% 3 of 6 | 46.2% | $0 |
Swipe the table sideways for more columns.
Ranks follow the score as shown, so equal numbers share a rank. Each model runs at its provider's default reasoning setting, one deck per task. The rating comes from head-to-head comparisons, so it keeps separating models even when several make presentable decks; ranges are 95% intervals from resampling the comparisons.
See the decks
Every model's deck for every public task, as rendered from the slides it returned. Open a deck to see each slide, the judges' rating and what stopped it being presentable.
Hearthside Coffee: Manchester store investment review · for the board of directors (non-executives and the chief executive)
GPT-6 Astra
✓ Presentable
GPT-6 Sol
✗ 1 issue
GPT-6.1 Sol
✓ Presentable
Claude Opus 5.5
✓ Presentable
Claude Sonnet 5.5
✓ Presentable
Claude Fable 5.1
✗ 3 issues
GPT-6 Luna
✗ 1 issue
Grok 4.7
✓ Presentable
Muse Wonder
✗ 2 issues
Kimi K3
✗ 5 issues
Claude Haiku 5.5
✗ 2 issues
DeepSeek V4.1 Flash 1017
✗ 1 issue
Qwen3.8 Max (0902)
✗ 2 issues
minimax-m3.1-flash:free
✗ 5 issues
Gemini 4 Argon
✗ 2 issues
Muse Spark 1.3
✗ 2 issues
Gemini 3.8 Flash
✗ 2 issues
No usable deck
Mercury 3
✗ No deck
Gemini 3.1 Pro Preview
✗ 2 issues
Claude Haiku 4.5
✗ 2 issues
Gemini 3.5 Flash Lite
✗ 4 issues
Mistral Medium 3.5
✗ 1 issue
Mistral Large 4
✗ 3 issues
DeepSeek V4 Pro 0423
✗ 2 issues
No usable deck
GLM 5.3
✗ No deck
Lumio: Starter churn: what to fund in Q1 2026 · for the leadership team (chief executive, CFO, heads of product, sales and customer success)
GPT-6 Astra
✓ Presentable
GPT-6 Sol
✓ Presentable
GPT-6.1 Sol
✗ 2 issues
Claude Opus 5.5
✗ 1 issue
Claude Sonnet 5.5
✗ 1 issue
Claude Fable 5.1
✓ Presentable
GPT-6 Luna
✗ 1 issue
Grok 4.7
✗ 1 issue
Muse Wonder
✗ 2 issues
Kimi K3
✗ 1 issue
No usable deck
Claude Haiku 5.5
✗ No deck
DeepSeek V4.1 Flash 1017
✗ 2 issues
Qwen3.8 Max (0902)
✗ 3 issues
minimax-m3.1-flash:free
✗ 1 issue
Gemini 4 Argon
✗ 3 issues
Muse Spark 1.3
✗ 2 issues
Gemini 3.8 Flash
✗ 2 issues
Mercury 3
✗ 1 issue
Gemini 3.1 Pro Preview
✗ 1 issue
Claude Haiku 4.5
✗ 3 issues
Gemini 3.5 Flash Lite
✗ 3 issues
Mistral Medium 3.5
✗ 1 issue
Mistral Large 4
✗ 5 issues
DeepSeek V4 Pro 0423
✗ 4 issues
No usable deck
GLM 5.3
✗ No deck
Pantry Lane: Paid media: H1 2026 allocation · for the CMO and the finance director
GPT-6 Astra
✓ Presentable
GPT-6 Sol
✗ 1 issue
GPT-6.1 Sol
✗ 1 issue
Claude Opus 5.5
✗ 1 issue
Claude Sonnet 5.5
✓ Presentable
Claude Fable 5.1
✗ 3 issues
GPT-6 Luna
✗ 1 issue
Grok 4.7
✓ Presentable
Muse Wonder
✗ 2 issues
Kimi K3
✗ 1 issue
Claude Haiku 5.5
✓ Presentable
DeepSeek V4.1 Flash 1017
✗ 3 issues
Qwen3.8 Max (0902)
✓ Presentable
minimax-m3.1-flash:free
✗ 2 issues
Gemini 4 Argon
✗ 1 issue
Muse Spark 1.3
✗ 1 issue
Gemini 3.8 Flash
✗ 5 issues
Mercury 3
✗ 1 issue
Gemini 3.1 Pro Preview
✓ Presentable
Claude Haiku 4.5
✗ 2 issues
Gemini 3.5 Flash Lite
✗ 4 issues
Mistral Medium 3.5
✗ 1 issue
Mistral Large 4
✗ 2 issues
DeepSeek V4 Pro 0423
✗ 4 issues
No usable deck
GLM 5.3
✗ No deck
Northgate Homes: Responsive repairs: in-house team or renewed contract · for the board of trustees of a housing association, including two tenant board members
GPT-6 Astra
✗ 1 issue
GPT-6 Sol
✓ Presentable
GPT-6.1 Sol
✗ 1 issue
Claude Opus 5.5
✓ Presentable
Claude Sonnet 5.5
✓ Presentable
Claude Fable 5.1
✗ 2 issues
GPT-6 Luna
✗ 1 issue
Grok 4.7
✗ 1 issue
Muse Wonder
✓ Presentable
Kimi K3
✗ 1 issue
Claude Haiku 5.5
✗ 1 issue
DeepSeek V4.1 Flash 1017
✗ 2 issues
Qwen3.8 Max (0902)
✗ 1 issue
minimax-m3.1-flash:free
✗ 2 issues
Gemini 4 Argon
✗ 2 issues
Muse Spark 1.3
✗ 4 issues
Gemini 3.8 Flash
✗ 1 issue
Mercury 3
✗ 2 issues
Gemini 3.1 Pro Preview
✗ 3 issues
Claude Haiku 4.5
✗ 3 issues
Gemini 3.5 Flash Lite
✗ 2 issues
Mistral Medium 3.5
✗ 4 issues
Mistral Large 4
✗ 5 issues
DeepSeek V4 Pro 0423
✗ 4 issues
No usable deck
GLM 5.3
✗ No deck
Kestrel Logistics: On-time delivery: what is going wrong and what to fund · for the board and the company's three main investors
GPT-6 Astra
✓ Presentable
GPT-6 Sol
✓ Presentable
GPT-6.1 Sol
✓ Presentable
Claude Opus 5.5
✗ 2 issues
Claude Sonnet 5.5
✗ 1 issue
Claude Fable 5.1
✗ 1 issue
GPT-6 Luna
✗ 1 issue
Grok 4.7
✗ 1 issue
Muse Wonder
✗ 2 issues
Kimi K3
✗ 1 issue
Claude Haiku 5.5
✓ Presentable
DeepSeek V4.1 Flash 1017
✗ 2 issues
Qwen3.8 Max (0902)
✓ Presentable
No usable deck
minimax-m3.1-flash:free
✗ No deck
Gemini 4 Argon
✗ 3 issues
Muse Spark 1.3
✗ 3 issues
Gemini 3.8 Flash
✗ 4 issues
Mercury 3
✗ 2 issues
Gemini 3.1 Pro Preview
✗ 3 issues
Claude Haiku 4.5
✗ 2 issues
Gemini 3.5 Flash Lite
✗ 1 issue
Mistral Medium 3.5
✗ 3 issues
Mistral Large 4
✗ 4 issues
DeepSeek V4 Pro 0423
✗ 4 issues
No usable deck
GLM 5.3
✗ No deck
Wrenfield Borough Council: Library services: choosing how to save £350,000 · for the council's cabinet (elected members), meeting in public
GPT-6 Astra
✗ 1 issue
GPT-6 Sol
✗ 1 issue
GPT-6.1 Sol
✓ Presentable
Claude Opus 5.5
✗ 2 issues
Claude Sonnet 5.5
✗ 2 issues
Claude Fable 5.1
✓ Presentable
GPT-6 Luna
✗ 1 issue
Grok 4.7
✗ 2 issues
Muse Wonder
✗ 1 issue
Kimi K3
✗ 1 issue
Claude Haiku 5.5
✗ 1 issue
DeepSeek V4.1 Flash 1017
✗ 5 issues
Qwen3.8 Max (0902)
✗ 3 issues
No usable deck
minimax-m3.1-flash:free
✗ No deck
Gemini 4 Argon
✗ 1 issue
Muse Spark 1.3
✗ 4 issues
Gemini 3.8 Flash
✗ 2 issues
Mercury 3
✗ 1 issue
Gemini 3.1 Pro Preview
✗ 3 issues
Claude Haiku 4.5
✗ 4 issues
Gemini 3.5 Flash Lite
✗ 3 issues
Mistral Medium 3.5
✗ 3 issues
Mistral Large 4
✗ 3 issues
DeepSeek V4 Pro 0423
✗ 3 issues
No usable deck
GLM 5.3
✗ No deck
More from the results
Accurate is not the same as presentable
Share of tasks where the deck said only what the analysis supports, against the share that could also be presented as it is. The gap is layout and design.
AccuratePresentable
Muse Wonder
100.0% · 16.7%
GPT-6.1 Sol
100.0% · 50.0%
GPT-6 Astra
100.0% · 66.7%
Claude Opus 5.5
83.3% · 33.3%
GPT-6 Sol
83.3% · 50.0%
Claude Fable 5.1
66.7% · 33.3%
Claude Sonnet 5.5
66.7% · 50.0%
Grok 4.7
66.7% · 33.3%
Qwen3.8-Max (0902)
50.0% · 33.3%
Gemini 3.1 Pro Preview
50.0% · 16.7%
MiniMax M3.1 Flash
50.0% · 0.0%
Kimi K3
50.0% · 0.0%
GPT-6 Luna
50.0% · 0.0%
Claude Haiku 5.5
33.3% · 33.3%
DeepSeek-V4.1-Flash-1017
33.3% · 0.0%
Gemini 4 Argon
33.3% · 0.0%
Claude Haiku 4.5
16.7% · 0.0%
Gemini 3.8 Flash
16.7% · 0.0%
Muse Spark 1.3
16.7% · 0.0%
DeepSeek-V4-Pro (0423)
0.0% · 0.0%
Gemini 3.5 Flash-Lite
0.0% · 0.0%
Mercury 3
0.0% · 0.0%
Mistral Large 4
0.0% · 0.0%
Mistral Medium 3.5
0.0% · 0.0%
Why decks are not presentable
Share of tasks failing each check. A deck can fail several at once; any one means it needs fixing before it is shown.
| Model | Layout defects | Slides needing work | Draft figure quoted | Unsupported numbers | Unsupported claims | Caveat dropped | Misleading metric | Findings missing | Recommendation |
|---|---|---|---|---|---|---|---|---|---|
| GPT-6 AstraOpenAI | 0.0% | 33.3% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% |
| GPT-6 SolOpenAI | 33.3% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 16.7% | 0.0% |
| GPT-6.1 SolOpenAI | 16.7% | 50.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% |
| Claude Opus 5.5Anthropic | 16.7% | 66.7% | 0.0% | 0.0% | 0.0% | 16.7% | 0.0% | 0.0% | 0.0% |
| Claude Sonnet 5.5Anthropic | 0.0% | 33.3% | 0.0% | 0.0% | 0.0% | 16.7% | 0.0% | 0.0% | 16.7% |
| Claude Fable 5.1Anthropic | 66.7% | 50.0% | 33.3% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% |
| GPT-6 LunaOpenAI | 33.3% | 16.7% | 33.3% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 16.7% |
| Grok 4.7xAI | 16.7% | 16.7% | 16.7% | 0.0% | 0.0% | 16.7% | 0.0% | 0.0% | 16.7% |
| Muse WonderMeta | 83.3% | 66.7% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% |
| Kimi K3Moonshot AI | 66.7% | 16.7% | 0.0% | 0.0% | 33.3% | 0.0% | 0.0% | 16.7% | 33.3% |
| Claude Haiku 5.5Anthropic | 33.3% | 16.7% | 33.3% | 16.7% | 33.3% | 16.7% | 16.7% | 16.7% | 33.3% |
| DeepSeek-V4.1-Flash-1017DeepSeek | 83.3% | 66.7% | 50.0% | 0.0% | 16.7% | 16.7% | 0.0% | 0.0% | 16.7% |
| Qwen3.8-Max (0902)Alibaba | 33.3% | 50.0% | 33.3% | 0.0% | 16.7% | 16.7% | 0.0% | 0.0% | 0.0% |
| MiniMax M3.1 FlashMiniMax | 100.0% | 83.3% | 33.3% | 50.0% | 50.0% | 33.3% | 33.3% | 50.0% | 33.3% |
| Gemini 4 ArgonGoogle | 66.7% | 66.7% | 50.0% | 0.0% | 16.7% | 0.0% | 0.0% | 0.0% | 0.0% |
| Muse Spark 1.3Meta | 100.0% | 50.0% | 16.7% | 0.0% | 16.7% | 16.7% | 0.0% | 0.0% | 66.7% |
| Gemini 3.8 FlashGoogle | 83.3% | 33.3% | 50.0% | 16.7% | 50.0% | 0.0% | 0.0% | 0.0% | 33.3% |
| Mercury 3Inception | 33.3% | 33.3% | 33.3% | 16.7% | 16.7% | 16.7% | 16.7% | 16.7% | 83.3% |
| Gemini 3.1 Pro PreviewGoogle | 50.0% | 83.3% | 0.0% | 0.0% | 0.0% | 16.7% | 0.0% | 0.0% | 50.0% |
| Claude Haiku 4.5Anthropic | 50.0% | 66.7% | 50.0% | 0.0% | 33.3% | 0.0% | 0.0% | 16.7% | 50.0% |
| Gemini 3.5 Flash-LiteGoogle | 66.7% | 16.7% | 50.0% | 0.0% | 16.7% | 16.7% | 0.0% | 33.3% | 83.3% |
| Mistral Medium 3.5Mistral AI | 33.3% | 50.0% | 50.0% | 0.0% | 33.3% | 0.0% | 0.0% | 0.0% | 50.0% |
| Mistral Large 4Mistral AI | 66.7% | 100.0% | 16.7% | 33.3% | 50.0% | 16.7% | 0.0% | 0.0% | 83.3% |
| DeepSeek-V4-Pro (0423)DeepSeek | 66.7% | 83.3% | 33.3% | 0.0% | 66.7% | 16.7% | 0.0% | 16.7% | 66.7% |
Every task, every model
Whether each deck could be presented as it is, and if not, how many kinds of issue it had. Each square opens the deck.
| Model | Hearthside Coffee | Lumio | Pantry Lane | Northgate Homes | Kestrel Logistics | Wrenfield Borough Council | Presentable |
|---|---|---|---|---|---|---|---|
| GPT-6 Astra | ✓ | ✓ | ✓ | 1 | ✓ | 1 | 4 of 6 |
| GPT-6 Sol | 1 | ✓ | 1 | ✓ | ✓ | 1 | 3 of 6 |
| GPT-6.1 Sol | ✓ | 2 | 1 | 1 | ✓ | ✓ | 3 of 6 |
| Claude Sonnet 5.5 | ✓ | 1 | ✓ | ✓ | 1 | 2 | 3 of 6 |
| Claude Opus 5.5 | ✓ | 1 | 1 | ✓ | 2 | 2 | 2 of 6 |
| Claude Fable 5.1 | 3 | ✓ | 3 | 2 | 1 | ✓ | 2 of 6 |
| Grok 4.7 | ✓ | 1 | ✓ | 1 | 1 | 2 | 2 of 6 |
| Claude Haiku 5.5 | 2 | – | ✓ | 1 | ✓ | 1 | 2 of 6 |
| Qwen3.8 Max (0902) | 2 | 3 | ✓ | 1 | ✓ | 3 | 2 of 6 |
| Muse Wonder | 2 | 2 | 2 | ✓ | 2 | 1 | 1 of 6 |
| Gemini 3.1 Pro Preview | 2 | 1 | ✓ | 3 | 3 | 3 | 1 of 6 |
| GPT-6 Luna | 1 | 1 | 1 | 1 | 1 | 1 | 0 of 6 |
| Kimi K3 | 5 | 1 | 1 | 1 | 1 | 1 | 0 of 6 |
| DeepSeek V4.1 Flash 1017 | 1 | 2 | 3 | 2 | 2 | 5 | 0 of 6 |
| minimax-m3.1-flash:free | 5 | 1 | 2 | 2 | – | – | 0 of 6 |
| Gemini 4 Argon | 2 | 3 | 1 | 2 | 3 | 1 | 0 of 6 |
| Muse Spark 1.3 | 2 | 2 | 1 | 4 | 3 | 4 | 0 of 6 |
| Gemini 3.8 Flash | 2 | 2 | 5 | 1 | 4 | 2 | 0 of 6 |
| Mercury 3 | – | 1 | 1 | 2 | 2 | 1 | 0 of 6 |
| Claude Haiku 4.5 | 2 | 3 | 2 | 3 | 2 | 4 | 0 of 6 |
| Gemini 3.5 Flash Lite | 4 | 3 | 4 | 2 | 1 | 3 | 0 of 6 |
| Mistral Medium 3.5 | 1 | 1 | 1 | 4 | 3 | 3 | 0 of 6 |
| Mistral Large 4 | 3 | 5 | 2 | 5 | 4 | 3 | 0 of 6 |
| DeepSeek V4 Pro 0423 | 2 | 4 | 4 | 4 | 4 | 3 | 0 of 6 |
| GLM 5.3 | – | – | – | – | – | – | 0 of 6 |
✓ presentable as it is · a number: the kinds of issue that stopped it · – no usable deck.
How DeckBench works
- 1
The analysis
An invented company's memo with findings, a recommendation and caveats, two to four data tables, the audience, the decision and a brand kit.
- 2
The deck
The model returns every slide as positioned text, shapes, native charts and tables. We render it to PowerPoint and to images.
- 3
The checks
Code checks the layout and traces every number to the analysis; three judges check the story and rate each slide's design.
- 4
Head to head
On each task the judges compare decks in pairs and pick the one they would rather present. The votes become the rating.
The task
Turning an analysis into slides is one of the most common things people ask AI to do at work, and the risk is not ugly slides: it is a number that is not in the analysis, a caveat that falls off, or a draft figure that ends up in front of the board.
Each task is an invented company's analysis written for the benchmark, with planted traps: a caveat that must travel with its number, a figure the memo says not to quote, a metric it warns is misleading, and a finding that contradicts a senior stakeholder's view.
What makes a deck presentable
- Layout
- Nothing off the slide or colliding, body text at least 12 points, and text readable against its background. Text wraps from real font metrics, so overflow is measured.
- Numbers
- Every number on a slide, in a chart or in a table must come from the analysis; numbers code cannot match go to the judges.
- Story
- The key findings, the recommendation within the first two slides, caveats with their numbers, and no misleading metric used as evidence.
- Design
- Each slide rated polished, acceptable or needs work. A presentable deck has no slide that needs work.
Why a rating as well
Presentable is a bar: once the best models clear it on every task, it stops telling them apart. The head-to-head rating does not run out of room, because a deck only has to be preferred to another one. Each deck meets about six others per task; a 400-point gap means the higher-rated model's deck is preferred about ten times out of eleven.
What it measures
- Every number traced to the analysis
- The story: findings, recommendation and caveats
- Layout and legibility, checked by code
- Design quality, judged slide by slide
What it does not measure
- Your templates, data or brand guidelines
- Decks built in PowerPoint or Google Slides by an agent clicking through the app
- Speaking notes or delivery
Method
- Layout, legibility and numbers are checked by code
- Three judges from three vendors decide each judged check by majority
- No judge compares decks from its own vendor
- Two tasks are held back so the set can be refreshed
Checking the judge
A single judge was the strictest of three on unsupported claims (it flagged 24 material claims in a pilot where the other two flagged 12 and 1), and its own model led the pilot. So every judged check is decided by a majority of three judges: GPT-6.1 Sol, Claude Opus 5.5 and Gemini 3.1 Pro. In the pilot they agreed with each other on 97 to 100% of the checks on findings, numbers, recommendations and misleading metrics, and on 89 to 90% of decisions about whether a slide needs work. In head-to-head comparisons a judge never votes on a pair that includes its own vendor's deck. A sample of human ratings is still to come.
Failures
A model that made no usable deck on any task has no rating, so it is listed here instead of ranked. They are listed here so you can see why.
Not ranked
- GLM 5.3: no usable deck: reply cut off at the token limit on 6 of 6 tasks