Model Comparison
DeepSeek V4.1 Flash vs Claude Haiku 5.5
Too close to call: the best configurations, DeepSeek V4.1 Flash (max) and Claude Haiku 5.5 (max effort), are 31 Elo apart, and their 95% ranges overlap. Among 78 models, each ranked by its best configuration, DeepSeek V4.1 Flash is #14 and Claude Haiku 5.5 is #16. Cost per script: $0.017 for DeepSeek V4.1 Flash (max) and $0.103 for Claude Haiku 5.5 (max effort).
Last updated , from board data generated at 01:13 UTC.
Picks and Verdict
PickDeepSeek V4.1 Flash
DeepSeek V4.1 Flash (high)
Pick DeepSeek V4.1 Flash (high) for bulk drafts, since it costs the same $0.012 a script as Claude Haiku 5.5 (high) and rates 248 Elo higher.
PickClaude Haiku 5.5
Claude Haiku 5.5 (xhigh)
Pick Claude Haiku 5.5 (xhigh) when substance and visual cues come first, since it keeps most of Haiku's lead there for $0.029 a script.
Claude Haiku 5.5 only gets within range of DeepSeek V4.1 Flash's best at max effort, its slowest and most expensive setting. Claude Haiku 5.5 (max effort) and DeepSeek V4.1 Flash (max) are 31 Elo apart with overlapping 95% ranges, so the board can't order the two best settings, the ones that make DeepSeek the #14 model of 78 and Haiku the #16 model. The bill can: DeepSeek costs $0.017 a script against $0.103, and takes 2.4 minutes against more than twelve. Below the top it gets lopsided. DeepSeek V4.1 Flash (high) costs $0.012 and its range still overlaps Haiku's best. Claude Haiku 5.5 (high) costs the same and sits 248 Elo lower with a separate range. DeepSeek leads at every level name the two share, and even with thinking off it beats Haiku at low by 134 Elo for less money. Haiku's case is substance and visual cues, about three and five points higher at the top, plus a cleaner slop score. Claude Haiku 5.5 (xhigh) keeps most of that for $0.029. I'd still default to DeepSeek, and the open weights help.
Disclosure: Claude Haiku 5.5 comes from the same lab as one of the four judges, Claude Opus 5.5. The other three sit outside its lab, so it is never scored by its own lab alone (how the panel works). The judge analysis measures how much each judge favors its own lab.
How to Read the Writing Score, Writing Elo and 95% Ranges
Writing Score and Writing Elo. The Writing Score (0 to 100) is the weighted average of the nine metric scores: seven from the judges, Length Adherence from code and the Slop Score half from each. The Writing Elo is a rating from pairwise comparisons: on each script, each configuration's average across its drafts is compared with every other configuration's, and gaps within the draft-to-draft variation count as draws. Elo rewards winning often, so the two can put configurations in a different order.
Elo ranges. Each 95% range shows how far a configuration's Elo could move if the drafts had come out slightly differently. It comes from resampling every configuration's recorded draft scores on each reference script and ranking again. Comparing two separate ranges does not test the Elo gap between them: ranges far apart are strong evidence for the order, and overlapping ranges do not mean a tie. The reference scripts stay fixed, so the ranges say nothing about new scripts. The methodology page has the full description.
All Thinking Levels of Both Models
Every ranked configuration of DeepSeek V4.1 Flash and Claude Haiku 5.5, scored on the same 10 scripts by the same four judges.
- Each row is one configuration: the model at one thinking level. The dot is its Writing Elo, a rating from pairwise comparisons of the scores on each script, and the bar is its 95% range: how far that Elo could move if the drafts had come out slightly differently.
- When two bars overlap, the board cannot order those two configurations. Bars far apart are strong evidence for the order.
- The table under the chart lists the same configurations in the same order and adds the Writing Score (the 0 to 100 consensus of the judges and code), cost, time and configuration rank (the place among all configurations on the board).
Shaded rows are the configurations chosen for the metric comparison below.
| Configuration | Writing Elo | 95% Range | Writing Score | Cost / Script | Time / Script | Reasoning Tokens | Configuration Rank |
|---|---|---|---|---|---|---|---|
| DeepSeek V4.1 Flash (max) | 2040 | 1999 to 2076 | 81.0 | $0.017 | 2.4 min | 22,010 | #28 |
| Claude Haiku 5.5 (max effort) | 2009 | 1972 to 2044 | 80.8 | $0.103 | 12.6 min | N/A | #31 |
| DeepSeek V4.1 Flash (high) | 1971 | 1925 to 2010 | 80.3 | $0.012 | 1.1 min | 13,809 | #38 |
| Claude Haiku 5.5 (xhigh) | 1927 | 1883 to 1965 | 80.0 | $0.029 | 3.6 min | N/A | #41 |
| DeepSeek V4.1 Flash (low) | 1856 | 1815 to 1906 | 79.3 | $0.0077 | 0.7 min | 6,740 | #46 |
| Claude Haiku 5.5 (high) | 1723 | 1679 to 1760 | 78.0 | $0.012 | 1.5 min | N/A | #53 |
| DeepSeek V4.1 Flash (no thinking) | 1638 | 1588 to 1680 | 76.8 | $0.0037 | 0.3 min | 0 | #61 |
| Claude Haiku 5.5 (medium) | 1620 | 1572 to 1670 | 76.7 | $0.0067 | 0.8 min | N/A | #62 |
| Claude Haiku 5.5 (low) | 1504 | 1461 to 1550 | 75.2 | $0.0045 | 0.5 min | N/A | #74 |
Cost is one script at the API list price of the route used, uncached and without judging. Time is the average for one accepted script, retries included. N/A under Reasoning Tokens means the board has no separate reasoning count for that route.
Computed Findings Across Thinking Levels
Each finding is computed from the configurations above. Where two 95% ranges overlap, the board cannot order those two configurations.
- Best Configurations Head to Head. DeepSeek V4.1 Flash (max) at Elo 2040 (range 1999 to 2076) and Claude Haiku 5.5 (max effort) at Elo 2009 (range 1972 to 2044) have overlapping 95% ranges, so the board cannot separate the two best configurations.
- Lowest Thinking Levels Head to Head. At their lowest thinking levels, DeepSeek V4.1 Flash (no thinking) at Elo 1638 and $0.0037 per script leads Claude Haiku 5.5 (low) at Elo 1504 and $0.0045 per script by 134 Elo. Their 95% ranges do not overlap.
- Same Level Name, Side by Side. DeepSeek V4.1 Flash has the higher Writing Elo at every level name both models use: low, high, and max. The 95% ranges do not overlap at low and high. Level names come from each provider, so the same name does not promise the same reasoning budget.
- Cheapest Configuration Overlapping the Top One. DeepSeek V4.1 Flash (high), at $0.012 per script, is the cheapest configuration whose 95% range overlaps that of DeepSeek V4.1 Flash (max), at $0.017 per script: 1925 to 2010 against 1999 to 2076, so the board cannot separate them. DeepSeek V4.1 Flash (max) costs about 1.4 times as much per script.
- Cheapest DeepSeek V4.1 Flash Configuration Above Claude Haiku 5.5's Best. The cheapest DeepSeek V4.1 Flash configuration with a higher Writing Elo than Claude Haiku 5.5 (max effort), at Elo 2009 and $0.103 per script, is DeepSeek V4.1 Flash (max), at Elo 2040 and $0.017 per script. Their 95% ranges overlap, so the board cannot separate them. It costs less than the configuration it beats.
- Lowest to Highest Thinking Level. DeepSeek V4.1 Flash moves +402 Elo from no thinking to max, at about 4.6 times the cost per script. Claude Haiku 5.5 moves +505 Elo from low to max effort, at about 23 times the cost per script.
Writing Elo at Each Shared Level Name
| Level | DeepSeek V4.1 Flash | Claude Haiku 5.5 | Gap | 95% Ranges |
|---|---|---|---|---|
| low | 1856 | 1504 | +352 | Separate |
| high | 1971 | 1723 | +248 | Separate |
| max | 2040 | 2009 | +31 | Overlap |
Gap is DeepSeek V4.1 Flash minus Claude Haiku 5.5 at the same level name, using each model's highest-rated configuration at that level.
Two Chosen Configurations Compared
Pick one configuration of each model to compare their metric scores, consistency across drafts and per-script scores. Each menu starts on the model's best configuration.
Scores by Metric
DeepSeek V4.1 Flash (max) scores higher on five of the nine metrics and Claude Haiku 5.5 (max effort) on four. Weighted by each metric's share of the Writing Score, YouTube Best Practices does the most to put DeepSeek V4.1 Flash (max) ahead, with a 3.0-point lead on that metric. The weight beside each metric is its share of the Writing Score.
Consistency Across Drafts
| DeepSeek V4.1 Flash (max) | Claude Haiku 5.5 (max effort) | |
|---|---|---|
| Score spread across drafts (±) | 1.30 | 1.55 |
| Lowest script score | 79.7 Script 10 | 78.2 Script 10 |
| Highest script score | 82.4 Script 7 | 82.6 Script 4 |
| Range across scripts | 2.7 | 4.4 |
Spread is how much the Writing Score moves between a configuration's drafts, and lower is steadier. Green marks the better number in each row.
Scores by Script
Writing Score on each reference script. Gap is DeepSeek V4.1 Flash (max) minus Claude Haiku 5.5 (max effort), and the higher score is in bold.
| Script | DeepSeek V4.1 Flash (max) | Claude Haiku 5.5 (max effort) | Gap |
|---|---|---|---|
| Script 1 opinion / warning explainer | 80.8 | 80.3 | +0.5 |
| Script 2 news-analysis / skeptical explainer | 81.2 | 80.7 | +0.5 |
| Script 3 personal roadmap / opinion | 81.2 | 80.8 | +0.4 |
| Script 4 short explainer | 80.1 | 82.6 | -2.5 |
| Script 5 news-analysis / opinion explainer | 81.1 | 81.7 | -0.6 |
| Script 6 founder announcement / personal origin story | 80.8 | 81.1 | -0.3 |
| Script 7 personal technical walkthrough / agentic workflow case study | 82.4 | 80.0 | +2.4 |
| Script 8 product release announcement / personal observation | 80.7 | 81.3 | -0.6 |
| Script 9 career guide / hiring analysis | 82.2 | 81.5 | +0.7 |
| Script 10 engineering process walkthrough / presentation adaptation | 79.7 | 78.2 | +1.5 |
Scorecards and Thinking Pages
Each configuration name in the table above opens its own scorecard.
Compare With Another Model
Pick a second model. A pair with its own written match-up opens that page, and any other pair opens in the comparison tool.
Nearby Match-Ups by Best Writing Elo
Every configuration wrote five drafts for each of the same 10 briefs, from the same prompt. Generation was closed-book, with the prompt as the only source and no web search or other tools. Each judge scores every draft once: Claude Opus 5.5 at medium effort through the Claude Code subscription, GPT-6.1 Sol at medium reasoning through the Codex CLI, GLM 5.3 Flash through OpenRouter, and Jev 1.13, a typed judge from TypeSafe that answers each metric as a multiple-choice question. The channel's finished script is their reference for the voice, and it gets no score on the board itself.
← All comparisons