Puter Developer Blog
News and updates from the Puter Engineering Team
Muse Wonder Review: Asking When a Request Is Unclear
We tested Muse Wonder against Muse Spark 1.3 across 140 tool-calling runs to see when it asks for clarification, when it checks in before acting, and what each run costs.
Gemini 4 Argon Review: Writing More Directly
We tested Gemini 4 Argon's writing against Gemini 3.8 Flash across 140 runs to see how directly it answers, how closely it follows style rules, and how often it falls back on stock phrases.
GPT-6 Sol and Luna Review: Frontier Performance, Budget Pricing
OpenAI cut API prices for GPT-6 Sol and Luna by 50% versus GPT-5.6. We priced an agentic coding session on both generations, and against Claude Opus 5.5, to see what the cut adds up to.
Grok 4.7 Review: xAI's Self-Checking Model
xAI says Grok 4.7 checks its own work more carefully than Grok 4.6. We ran 30 graded calls against both models and found no measurable difference at default settings.
Claude Opus 5.5 Review: Fable-Level Results for Less
Anthropic says Claude Opus 5.5 costs 40% less to run than Opus 5. We broke down the price changes and priced an agentic coding session on both models to see how they add up.
Best Text to Speech APIs in 2026
We compared the best text to speech APIs of 2026 on voice quality, latency, pricing, and voice cloning to find the right fit for each use case.
Best ElevenLabs Alternatives (2026)
Discover the best ElevenLabs alternatives, including Puter.js, OpenAI TTS, Deepgram, Amazon Polly, Chatterbox, and Speechify. Compare features, pricing, and find the best fit for your project.
GPT Image 2.5 Review: Sharper Details and More Precise Editing
OpenAI just released GPT Image 2.5 Flare and Sunburst, with sharper details, more precise editing, and faster generation. We look at what changed and put them to the test.
Using Blender with ChatGPT: Build 3D Models from a Prompt
How to use Blender with ChatGPT to build 3D models from a prompt. We go over the ways ChatGPT can work with Blender, then show how to do it in the browser without installing Blender.
Mercury 2.5 Review: Fastest Model for Latency Sensitive Workloads
Inception just released Mercury 2.5, their latest diffusion LLM with over 1,000 tokens per second. We tested what that speed actually looks like in practice, measuring both raw throughput and time to first token against Mercury 2, Claude Haiku 4.5 and Gemini 3.5 Flash-Lite.
GPT-6 Astra Review: A Taste of AGI
OpenAI just released GPT-6 Astra, which it calls the world's most intelligent and aligned model, with saturating scores on ARC-AGI-3, FrontierMath, and ExploitBench. We look at what is actually new and test one of its claims ourselves.
Gemini 3.8 Flash Review: A Diligent Model That Works Harder
Gemini 3.8 Flash matches Gemini 3.7 Flash on price but posts better benchmarks, which Google credits to the model being more diligent. We looked at the performance numbers and the diligence claim, and tested it ourselves.
Claude Fable 5.1 Review: Better Results with Less Effort
Claude Fable 5.1 is out. We put together a cost estimate of what its lower-effort, lower-cost design means for coding work.
Apodex 1.1 Review: Research With a Built-In Verification Step
Apodex says its Statement Review step independently checks key claims before delivery. We fed Apodex 1.1 and Apodex 1.1 Mini a prompt with a deliberately wrong gross margin figure and watched what they did with it.
GLM-5.3 Review: Stronger Coding With Fewer Tokens
Z.ai says GLM-5.3 beats GLM-5.2 with fewer output tokens. We ran 8 coding tasks against both models via Puter.js at max reasoning effort and measured the actual token spend.
Gemini 3.7 Flash Review: Google's Workhorse Model
Google says Gemini 3.7 Flash follows instructions with greater fidelity and needs less manual oversight than 3.6 Flash. We ran 12 single-turn tasks against both models through Puter.js and checked the outputs against binary pass criteria.
Grok 4.6 Review: xAI's Stronger Visual and Interactive Model
xAI says Grok 4.6 produces stronger first passes on visual and interactive projects than Grok 4.5. We built the same app with both and graded what came out.
Introducing BytePlus Models “Seedance, Seedream, and Seed” on Puter.js in Collaboration With the BytePlus Team
BytePlus, ByteDance's model platform, is now integrated directly with Puter, bringing the Seed, Seedream, and Seedance families to Puter.js, with new models added as they launch.
Announcing Puter AI Builder
We're launching AI Builder, a tool that lets you build apps using plain language, powered by Puter.js. Apps are secure by default, cost nothing for developers to run, and publish to their own link in one click.
Kimi K3 Review: Benchmarks, Pricing, and a Visual Coding Test
We reviewed Kimi K3, Moonshot AI's 2.8T open-weight model, a Fable 5-level model with Sonnet 5 pricing. Reported benchmarks vs independent numbers, the $3/$15 pricing math, and a hands-on visual coding test.