GPT-6.2 Astra just came out. Time for the classic pelican test.
Same prompt for GPT-6.2 Astra, GPT-6.2 Sol, Claude Opus 5.5 and Claude Fable 5.2.
Full prompt, every output, token count and cost 👇 It's 1 of 6 real jobs in this eval.
New models get the same jobs as they come out. Follow @monkeys_im so you don't miss the next one.
7
Watch every orbit play, with the full prompt and the cost and time of each run: monkeys.im/monkeys/evals/bes…
5
GPT-6.2 Astra just came out. Did Astra finally learn to walk?
Same walk-cycle prompt for all four.
GPT-6.2 Astra vs GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.2.
Full prompt, every output, token count and cost 👇 It's 1 of 6 real jobs in this eval.
New models get the same jobs as they come out. Follow @monkeys_im so you don't miss the next one.
8
Watch every walk play, with the full prompt and the cost and time of each run: monkeys.im/monkeys/evals/bes…
12
GPT-6.2 Astra just came out. Did Astra finally learn to walk?
Same walk-cycle prompt for all four.
GPT-6.2 Astra vs GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.2.
Full prompt, every output, token count and cost 👇 It's 1 of 6 real jobs in this eval.
14
GPT-6.1 Terra vs Qwen4 Max, Claude Sonnet 5.5 and Grok 4.7. Can they walk?
Same prompt: a person walking across the screen in a natural, looping walk cycle. Costs ran from $0.055 to $2.81.
It's 1 of 6 real jobs in this eval. Every output, token count and cost is public 👇
Watch every walk play, with the cost and time of each run: monkeys.im/monkeys/evals/bes…
1
203
New models get the same jobs as they come out. Follow @monkeys_im so you don't miss the next one.
204
GPT-6.2 Sol vs GPT-6.1 Sol, Gemini 4 Carbon and Claude Opus 5.5.
Same prompt: a miniature fishing harbor in a sunshower, in 3D, camera slowly orbiting. Costs ran from $0.28 to $7.72.
It's 1 of 6 real jobs in this eval. Every output, token count and cost is public 👇
Orbit every harbor yourself, with the cost and time of each run: monkeys.im/monkeys/evals/bes…
1
248
New models get the same jobs as they come out. Follow @monkeys_im so you don't miss the next one.
269
GPT-6.2 Sol is out today. Can it walk?
Same prompt for all four: a person walking across the screen in a natural, looping walk cycle.
GPT-6.2 Sol vs GPT-6.1 Sol, Claude Opus 5.5 and Gemini 4 Carbon.
Which walk looks the most natural?
Watch every walk play, with the cost and time of each run: monkeys.im/monkeys/evals/bes…
1
115
New models get the same jobs as they come out. Follow @monkeys_im so you don't miss the next one.
70
GPT-6.2 Sol is out today. How much better is it than GPT-6.1 Sol?
Same two animation jobs. Left: 6.1. Right: 6.2.
Top: a pelican riding a bicycle around Saturn's rings, SVG only.
Bottom: a person walking across the screen in a natural looping walk cycle.
Which upgrade do you see?
Watch both animations play, with the cost and time of every run, plus more Pro LLMs on the same jobs: monkeys.im/monkeys/evals/bes…
1
71
New models get the same jobs as they come out. Follow @monkeys_im so you don't miss the next one.
30
Same prompt. Same pelican. Same Saturn.
4 budget AI models, one job: animate a pelican riding a bicycle around Saturn's rings, in pure SVG.
GPT-6 Luna: $0.015, 8 min
GLM 5.3 Flash: $0.032, 10 min
Claude Haiku 5.5: $0.051, 8.5 min
DeepSeek V4.1 Flash: $0.71, 55 min
DeepSeek cost 47x more than Luna. Which one did it best? 👇
1
3,815
All 4 outputs side by side, plus the other 9 jobs, with every token count and cost: monkeys.im/monkeys/evals/bes…
1
48
Same 10 jobs, every model. Follow @monkeys_im so you don't miss the next match-up.
2
