GPT-6 Astra is the most creative AI model right now, with the top score on EQ-Bench's creative writing test. GPT-6 Sol comes close for 25% of the cost, and Grok 4.7 writes fastest.
GPT-6 Astra is the most creative AI model right now, with the top score on EQ-Bench's creative writing test. GPT-6 Sol comes close for 25% of the cost, and Grok 4.7 writes fastest.
21 models · Updated September 2026 · Data from EQ-Bench and Artificial Analysis
GPT-6 Astra
OpenAI · Medium effort
The highest creative writing score of the models we compared.
GPT-6 Sol
OpenAI · Extra high effort
The cheapest model within 200 points of the leader, at 25% of GPT-6 Astra's cost.
Grok 4.7
SpaceXAI · Extra high effort
The quickest full answer from a model scoring 1973 or more.
GPT-6 Astra
OpenAI · Medium effort
The highest creative writing score of the models we compared.
GPT-6 Sol
OpenAI · Extra high effort
The cheapest model within 200 points of the leader, at 25% of GPT-6 Astra's cost.
Grok 4.7
SpaceXAI · Extra high effort
The quickest full answer from a model scoring 1973 or more.
Each job favors a different model. Pick the one closest to yours.
Fiction, storytelling and scripts
Use
GPT-6 Astra
OpenAI · Medium effort
The top creative writing score of the models we compared, at $95.23 per 1,000 pieces.
Fiction, storytelling and scripts
Use GPT-6 Astra
The top creative writing score of the models we compared, at $95.23 per 1,000 pieces.
Blog posts and marketing copy at volume
Use GPT-6 Sol
The cheapest model within 200 points of the leader: $24.16 per 1,000 pieces, 25% of what GPT-6 Astra costs.
Brainstorming in a live chat
Use Grok 4.7
A full answer in about 12 seconds, so ideas keep coming while you think. It scores 2007.
Staying with Claude
Use Claude Fable 5.1
Claude's highest creative writing score, at 2162. Claude Opus 5.5 scores 2050 for $65.68 per 1,000 pieces against $149.
Open weights you can run yourself
Use GLM-5.3
An open-weight model with a creative writing score of 2075, close to the closed leaders, at $45.94 per 1,000 pieces through the API.
Making AI drafts sound like a person
Our free humanizer rewrites stiff AI text, removing stock phrases and uneven rhythm.
Each dot is a model. Higher scores better, further left is cheaper, so the best deals sit top left. The cost axis is logarithmic: every step left is a big saving. Hover a dot for its numbers.
Sort by the pillar you care about, and set your monthly volume to see the bill. Costs include each model's thinking tokens at the setting shown.
GPT-6 Astra, narrowly. On EQ-Bench Creative Writing v3 it rates 2173, ahead of Claude Fable 5.1 at 2162 and GPT-6 Sol at 2125.
The benchmark has a judge model compare short stories head to head and turns the results into an Elo rating. An 11-point lead means the leader wins about 52% of those comparisons, so the top few are close enough that your own taste should decide.
EQ-Bench also counts how often each model reaches for stock AI phrases, its slop score (lower is better). Among current models, Claude Fable 5.1 (8.16), GPT-6 Astra (8.41) and GLM-5.3 (8.42) use the fewest. Gemini 3.1 Pro uses the most, at 30.03.
A low slop score means less editing before a draft reads naturally. For the rest, a humanizer pass removes the most common tells.
EQ-Bench scores short fiction, not articles or ads, so treat it as a guide for business writing. The same top models are a sensible start: GPT-6 Astra for your best work, GPT-6 Sol for volume. A quick test on your own brief will tell you more than any leaderboard.
Creative writing is the Elo rating from EQ-Bench Creative Writing v3, checked September 2026. A judge model reads each model's short stories and compares them pair by pair. Models EQ-Bench hasn't tested aren't on this page.
Cost starts from a typical piece of 10,000 tokens, 60% your brief and context and 40% the writing. We scale the writing part by how much each model actually produced, thinking included, at its everyday setting in Artificial Analysis's tests.
Speed is the median time to a full answer at each model's strongest reasoning setting that answers within 45 seconds. The creative score doesn't change with that setting, so the Max effort switch only changes cost and speed here.
Best value is the cheapest model within 200 points of the leader, and fastest is the quickest full answer among the same group. Neither pick goes to a model its maker has replaced.
GPT-6 Astra. It has the highest EQ-Bench Creative Writing v3 score of the models we compared (2173) and costs about $95.23 per 1,000 pieces. GPT-6 Sol is close behind at 2125 for $24.16.
GPT-6 Astra, for the same reason: EQ-Bench's creative writing test is built from short stories, and it leads at 2173. If you want to write interactively, Grok 4.7 answers in about 12 seconds.
Claude Fable 5.1, with a creative writing score of 2162, the highest of the Claude models EQ-Bench has tested.
Less so than the leaders. Gemini 3.8 Flash, Google's best-scoring current model here, rates 1748 against 2173 for GPT-6 Astra, and it uses more stock phrases (slop score 22.6).
Yes. EQ-Bench Creative Writing v3 is the one we use: a judge model compares short stories from each model head to head and produces an Elo rating, plus measures of repetitive phrasing. Treat it as a guide and test the top few on your own writing.
Answer three questions and get a pick for each part of your workload, with the monthly cost.