Simon Willison’s Weblog

Gemini 3.1 Pro. The first in the Gemini 3.1 series, priced the same as Gemini 3 Pro ($2/million input, $12/million output under 200,000 tokens, $4/$18 for 200,000 to 1,000,000). That's less than half the price of Claude Opus 4.6 with very similar benchmark scores to that model.

They boast about its improved SVG animation performance compared to Gemini 3 Pro in the announcement!

I tried "Generate an SVG of a pelican riding a bicycle" in Google AI Studio and it thought for 323.9 seconds (thinking trace here) before producing this one:

It's good to see the legs clearly depicted on both sides of the frame (should satisfy Elon), the fish in the basket is a nice touch and I appreciated this comment in the SVG code:

<!-- Black Flight Feathers on Wing Tip -->
<path d="M 420 175 C 440 182, 460 187, 470 190 C 450 210, 430 208, 410 198 Z" fill="#374151" />

I've added the two new model IDs gemini-3.1-pro-preview and gemini-3.1-pro-preview-customtools to my llm-gemini plugin for LLM. That "custom tools" one is described here - apparently it may provide better tool performance than the default model in some situations.

The model appears to be incredibly slow right now - it took 104s to respond to a simple "hi" and a few of my other tests met "Error: This model is currently experiencing high demand. Spikes in demand are usually temporary. Please try again later." or "Error: Deadline expired before operation could complete" errors. I'm assuming that's just teething problems on launch day.

It sounds like last week's Deep Think release was our first exposure to the 3.1 family:

Last week, we released a major update to Gemini 3 Deep Think to solve modern challenges across science, research and engineering. Today, we’re releasing the upgraded core intelligence that makes those breakthroughs possible: Gemini 3.1 Pro.

Update: In What happens if AI labs train for pelicans riding bicycles? last November I said:

If a model finally comes out that produces an excellent SVG of a pelican riding a bicycle you can bet I’m going to test it on all manner of creatures riding all sorts of transportation devices.

Google's Gemini Lead Jeff Dean tweeted this video featuring an animated pelican riding a bicycle, plus a frog on a penny-farthing and a giraffe driving a tiny car and an ostrich on roller skates and a turtle kickflipping a skateboard and a dachshund driving a stretch limousine.

I've been saying for a while that I wish AI labs would highlight things that their new models can do that their older models could not, so top marks to the Gemini team for this video.

Update 2: I used llm-gemini to run my more detailed Pelican prompt, with this result:

From the SVG comments:

<!-- Pouch Gradient (Breeding Plumage: Red to Olive/Green) -->
...
<!-- Neck Gradient (Breeding Plumage: Chestnut Nape, White/Yellow Front) -->

# 19th February 2026, 5:58 pm / gemini, llm, pelican-riding-a-bicycle, ai, llms, llm-release, google, generative-ai, svg

Introducing Claude Sonnet 4.6 (via) Sonnet 4.6 is out today, and Anthropic claim it offers similar performance to November's Opus 4.5 while maintaining the Sonnet pricing of $3/million input and $15/million output tokens (the Opus models are $5/$25). Here's the system card PDF.

Sonnet 4.6 has a "reliable knowledge cutoff" of August 2025, compared to Opus 4.6's May 2025 and Haiku 4.5's February 2025. Both Opus and Sonnet default to 200,000 max input tokens but can stretch to 1 million in beta and at a higher cost.

I just released llm-anthropic 0.24 with support for both Sonnet 4.6 and Opus 4.6. Claude Code did most of the work - the new models had a fiddly amount of extra details around adaptive thinking and no longer supporting prefixes, as described in Anthropic's migration guide.

Here's what I got from:

uvx --with llm-anthropic llm 'Generate an SVG of a pelican riding a bicycle' -m claude-sonnet-4.6

The SVG comments include:

<!-- Hat (fun accessory) -->

I tried a second time and also got a top hat. Sonnet 4.6 apparently loves top hats!

For comparison, here's the pelican Opus 4.5 drew me in November:

And here's Anthropic's current best pelican, drawn by Opus 4.6 on February 5th:

Opus 4.6 produces the best pelican beak/pouch. I do think the top hat from Sonnet 4.6 is a nice touch though.

# 17th February 2026, 11:58 pm / llm, anthropic, claude, llm-pricing, ai, llms, llm-release, generative-ai, pelican-riding-a-bicycle, claude-code

Qwen3.5: Towards Native Multimodal Agents. Alibaba's Qwen just released the first two models in the Qwen 3.5 series - one open weights, one proprietary. Both are multi-modal for vision input.

The open weight one is a Mixture of Experts model called Qwen3.5-397B-A17B. Interesting to see Qwen call out serving efficiency as a benefit of that architecture:

Built on an innovative hybrid architecture that fuses linear attention (via Gated Delta Networks) with a sparse mixture-of-experts, the model attains remarkable inference efficiency: although it comprises 397 billion total parameters, just 17 billion are activated per forward pass, optimizing both speed and cost without sacrificing capability.

It's 807GB on Hugging Face, and Unsloth have a collection of smaller GGUFs ranging in size from 94.2GB 1-bit to 462GB Q8_K_XL.

I got this pelican from the OpenRouter hosted model (transcript):

The proprietary hosted model is called Qwen3.5 Plus 2026-02-15, and is a little confusing. Qwen researcher Junyang Lin says:

Qwen3-Plus is a hosted API version of 397B. As the model natively supports 256K tokens, Qwen3.5-Plus supports 1M token context length. Additionally it supports search and code interpreter, which you can use on Qwen Chat with Auto mode.

Here's its pelican, which is similar in quality to the open weights model:

# 17th February 2026, 4:30 am / vision-llms, ai, qwen, llms, ai-in-china, llm-release, generative-ai, openrouter, pelican-riding-a-bicycle

Introducing GPT‑5.3‑Codex‑Spark. OpenAI announced a partnership with Cerebras on January 14th. Four weeks later they're already launching the first integration, "an ultra-fast model for real-time coding in Codex".

Despite being named GPT-5.3-Codex-Spark it's not purely an accelerated alternative to GPT-5.3-Codex - the blog post calls it "a smaller version of GPT‑5.3-Codex" and clarifies that "at launch, Codex-Spark has a 128k context window and is text-only."

I had some preview access to this model and I can confirm that it's significantly faster than their other models.

Here's what that speed looks like running in Codex CLI:

That was the "Generate an SVG of a pelican riding a bicycle" prompt - here's the rendered result:

Compare that to the speed of regular GPT-5.3 Codex medium:

Significantly slower, but the pelican is a lot better:

What's interesting about this model isn't the quality though, it's the speed. When a model responds this fast you can stay in flow state and iterate with the model much more productively.

I showed a demo of Cerebras running Llama 3.1 70 B at 2,000 tokens/second against Val Town back in October 2024. OpenAI claim 1,000 tokens/second for their new model, and I expect it will prove to be a ferociously useful partner for hands-on iterative coding sessions.

It's not yet clear what the pricing will look like for this new model.

# 12th February 2026, 9:16 pm / inference-speed, openai, cerebras, pelican-riding-a-bicycle, ai, llms, llm-release, codex-cli, generative-ai

Gemini 3 Deep Think (via) New from Google. They say it's "built to push the frontier of intelligence and solve modern challenges across science, research, and engineering".

It drew me a really good SVG of a pelican riding a bicycle! I think this is the best one I've seen so far - here's my previous collection.

(And since it's an FAQ, here's my answer to What happens if AI labs train for pelicans riding bicycles?)

Since it did so well on my basic Generate an SVG of a pelican riding a bicycle I decided to try the more challenging version as well:

Generate an SVG of a California brown pelican riding a bicycle. The bicycle must have spokes and a correctly shaped bicycle frame. The pelican must have its characteristic large pouch, and there should be a clear indication of feathers. The pelican must be clearly pedaling the bicycle. The image should show the full breeding plumage of the California brown pelican.

Here's what I got:

# 12th February 2026, 6:12 pm / gemini, llm-reasoning, pelican-riding-a-bicycle, ai, llms, llm-release, google, generative-ai

GLM-5: From Vibe Coding to Agentic Engineering (via) This is a huge new MIT-licensed model: 754B parameters and 1.51TB on Hugging Face twice the size of GLM-4.7 which was 368B and 717GB (4.5 and 4.6 were around that size too).

It's interesting to see Z.ai take a position on what we should call professional software engineers building with LLMs - I've seen Agentic Engineering show up in a few other places recently. most notable from Andrej Karpathy and Addy Osmani.

I ran my "Generate an SVG of a pelican riding a bicycle" prompt through GLM-5 via OpenRouter and got back a very good pelican on a disappointing bicycle frame:

# 11th February 2026, 6:56 pm / pelican-riding-a-bicycle, ai, ai-in-china, llms, llm-release, vibe-coding, ai-assisted-programming, generative-ai, definitions, openrouter, glm

Kimi K2.5: Visual Agentic Intelligence (via) Kimi K2 landed in July as a 1 trillion parameter open weight LLM. It was joined by Kimi K2 Thinking in November which added reasoning capabilities. Now they've made it multi-modal: the K2 models were text-only, but the new 2.5 can handle image inputs as well:

Kimi K2.5 builds on Kimi K2 with continued pretraining over approximately 15T mixed visual and text tokens. Built as a native multimodal model, K2.5 delivers state-of-the-art coding and vision capabilities and a self-directed agent swarm paradigm.

The "self-directed agent swarm paradigm" claim there means improved long-sequence tool calling and training on how to break down tasks for multiple agents to work on at once:

For complex tasks, Kimi K2.5 can self-direct an agent swarm with up to 100 sub-agents, executing parallel workflows across up to 1,500 tool calls. Compared with a single-agent setup, this reduces execution time by up to 4.5x. The agent swarm is automatically created and orchestrated by Kimi K2.5 without any predefined subagents or workflow.

I used the OpenRouter Chat UI to have it "Generate an SVG of a pelican riding a bicycle", and it did quite well:

As a more interesting test, I decided to exercise the claims around multi-agent planning with this prompt:

I want to build a Datasette plugin that offers a UI to upload files to an S3 bucket and stores information about them in a SQLite table. Break this down into ten tasks suitable for execution by parallel coding agents.

Here's the full response. It produced ten realistic tasks and reasoned through the dependencies between them. For comparison here's the same prompt against Claude Opus 4.5 and against GPT-5.2 Thinking.

The Hugging Face repository is 595GB. The model uses Kimi's janky "modified MIT" license, which adds the following clause:

Our only modification part is that, if the Software (or any derivative works thereof) is used for any of your commercial products or services that have more than 100 million monthly active users, or more than 20 million US dollars (or equivalent in other currencies) in monthly revenue, you shall prominently display "Kimi K2.5" on the user interface of such product or service.

Given the model's size, I expect one way to run it locally would be with MLX and a pair of $10,000 512GB RAM M3 Ultra Mac Studios. That setup has been demonstrated to work with previous trillion parameter K2 models.

# 27th January 2026, 3:07 pm / vision-llms, ai-agents, llm-tool-use, pelican-riding-a-bicycle, ai, ai-in-china, llms, moonshot, kimi, parallel-agents, hugging-face, janky-licenses, llm-release

Introducing GPT-5.2-Codex. The latest in OpenAI's Codex family of models (not the same thing as their Codex CLI or Codex Cloud coding agent tools).

GPT‑5.2-Codex is a version of GPT‑5.2⁠ further optimized for agentic coding in Codex, including improvements on long-horizon work through context compaction, stronger performance on large code changes like refactors and migrations, improved performance in Windows environments, and significantly stronger cybersecurity capabilities.

As with some previous Codex models this one is available via their Codex coding agents now and will be coming to the API "in the coming weeks". Unlike previous models there's a new invite-only preview process for vetted cybersecurity professionals for "more permissive models".

I've been very impressed recently with GPT 5.2's ability to tackle multi-hour agentic coding challenges. 5.2 Codex scores 64% on the Terminal-Bench 2.0 benchmark that GPT-5.2 scored 62.2% on. I'm not sure how concrete that 1.8% improvement will be!

I didn't hack API access together this time (see previous attempts), instead opting to just ask Codex CLI to "Generate an SVG of a pelican riding a bicycle" while running the new model (effort medium). Here's the transcript in my new Codex CLI timeline viewer, and here's the pelican it drew:

# 19th December 2025, 5:21 am / gpt-codex, openai, pelican-riding-a-bicycle, ai, llms, llm-release, codex-cli, generative-ai

Devstral 2. Two new models from Mistral today: Devstral 2 and Devstral Small 2 - both focused on powering coding agents such as Mistral's newly released Mistral Vibe which I wrote about earlier today.

• Devstral 2: SOTA open model for code agents with a fraction of the parameters of its competitors and achieving 72.2% on SWE-bench Verified.
• Up to 7x more cost-efficient than Claude Sonnet at real-world tasks.

Devstral 2 is a 123B model released under a janky license - it's "modified MIT" where the modification is:

You are not authorized to exercise any rights under this license if the global consolidated monthly revenue of your company (or that of your employer) exceeds $20 million (or its equivalent in another currency) for the preceding month. This restriction in (b) applies to the Model and any derivatives, modifications, or combined works based on it, whether provided by Mistral AI or by a third party. [...]

Mistral Small 2 is under a proper Apache 2 license with no weird strings attached. It's a 24B model which is 51.6GB on Hugging Face and should quantize to significantly less.

I tried out the larger model via my llm-mistral plugin like this:

llm install llm-mistral
llm mistral refresh
llm -m mistral/devstral-2512 "Generate an SVG of a pelican riding a bicycle"

For a ~120B model that one is pretty good!

Here's the same prompt with -m mistral/labs-devstral-small-2512 for the API hosted version of Devstral Small 2:

Again, a decent result given the small parameter size. For comparison, here's what I got for the 24B Mistral Small 3.2 earlier this year.

# 9th December 2025, 11:58 pm / llm-release, mistral, generative-ai, ai, janky-licenses, llms, llm, pelican-riding-a-bicycle

Introducing Mistral 3. Four new models from Mistral today: three in their "Ministral" smaller model series (14B, 8B, and 3B) and a new Mistral Large 3 MoE model with 675B parameters, 41B active.

All of the models are vision capable, and they are all released under an Apache 2 license.

I'm particularly excited about the 3B model, which appears to be a competent vision-capable model in a tiny ~3GB file.

Xenova from Hugging Face got it working in a browser:

@MistralAI releases Mistral 3, a family of multimodal models, including three start-of-the-art dense models (3B, 8B, and 14B) and Mistral Large 3 (675B, 41B active). All Apache 2.0! 🤗

Surprisingly, the 3B is small enough to run 100% locally in your browser on WebGPU! 🤯

You can try that demo in your browser, which will fetch 3GB of model and then stream from your webcam and let you run text prompts against what the model is seeing, entirely locally.

Mistral's API hosted versions of the new models are supported by my llm-mistral plugin already thanks to the llm mistral refresh command:

$ llm mistral refresh
Added models: ministral-3b-2512, ministral-14b-latest, mistral-large-2512, ministral-14b-2512, ministral-8b-2512

I tried pelicans against all of the models. Here's the best one, from Mistral Large 3:

And the worst from Ministral 3B:

# 2nd December 2025, 5:30 pm / vision-llms, llm-release, mistral, llm, generative-ai, ai, llms

DeepSeek-V3.2 (via) Two new open weight (MIT licensed) models from DeepSeek today: DeepSeek-V3.2 and DeepSeek-V3.2-Speciale, both 690GB, 685B parameters. Here's the PDF tech report.

DeepSeek-V3.2 is DeepSeek's new flagship model, now running on chat.deepseek.com.

The difference between the two new models is best explained by this paragraph from the technical report:

DeepSeek-V3.2 integrates reasoning, agent, and human alignment data distilled from specialists, undergoing thousands of steps of continued RL training to reach the final checkpoints. To investigate the potential of extended thinking, we also developed an experimental variant, DeepSeek-V3.2-Speciale. This model was trained exclusively on reasoning data with a reduced length penalty during RL. Additionally, we incorporated the dataset and reward method from DeepSeekMath-V2 (Shao et al., 2025) to enhance capabilities in mathematical proofs.

I covered DeepSeek-Math-V2 last week. Like that model, DeepSeek-V3.2-Speciale also scores gold on the 2025 International Mathematical Olympiad so beloved of model training teams!

I tried both models on "Generate an SVG of a pelican riding a bicycle" using the chat feature of OpenRouter. DeepSeek V3.2 produced this very short reasoning chain:

Let's assume the following:

Wheel radius: 40
Distance between wheel centers: 180
Seat height: 60 (above the rear wheel center)
Handlebars: above the front wheel, extending back and up.

We'll set the origin at the center of the rear wheel.

We'll create the SVG with a viewBox that fits the entire drawing.

Let's start by setting up the SVG.

Followed by this illustration:

Here's what I got from the Speciale model, which thought deeply about the geometry of bicycles and pelicans for a very long time (at least 10 minutes) before spitting out this result:

# 1st December 2025, 11:56 pm / llm-release, openrouter, generative-ai, deepseek, ai, ai-in-china, llms, llm-reasoning, pelican-riding-a-bicycle

deepseek-ai/DeepSeek-Math-V2. New on Hugging Face, a specialist mathematical reasoning LLM from DeepSeek. This is their entry in the space previously dominated by proprietary models from OpenAI and Google DeepMind, both of which achieved gold medal scores on the International Mathematical Olympiad earlier this year.

We now have an open weights (Apache 2 licensed) 685B, 689GB model that can achieve the same. From the accompanying paper:

DeepSeekMath-V2 demonstrates strong performance on competition mathematics. With scaled test-time compute, it achieved gold-medal scores in high-school competitions including IMO 2025 and CMO 2024, and a near-perfect score on the undergraduate Putnam 2024 competition.

# 27th November 2025, 3:59 pm / llm-release, llm-reasoning, deepseek, ai, ai-in-china, llms, mathematics, generative-ai

Building more with GPT-5.1-Codex-Max (via) Hot on the heels of yesterday's Gemini 3 Pro release comes a new model from OpenAI called GPT-5.1-Codex-Max.

(Remember when GPT-5 was meant to bring in a new era of less confusing model names? That didn't last!)

It's currently only available through their Codex CLI coding agent, where it's the new default model:

Starting today, GPT‑5.1-Codex-Max will replace GPT‑5.1-Codex as the default model in Codex surfaces. Unlike GPT‑5.1, which is a general-purpose model, we recommend using GPT‑5.1-Codex-Max and the Codex family of models only for agentic coding tasks in Codex or Codex-like environments.

It's not available via the API yet but should be shortly.

The timing of this release is interesting given that Gemini 3 Pro appears to have aced almost all of the benchmarks just yesterday. It's reminiscent of the period in 2024 when OpenAI consistently made big announcements that happened to coincide with Gemini releases.

OpenAI's self-reported SWE-Bench Verified score is particularly notable: 76.5% for thinking level "high" and 77.9% for the new "xhigh". That was the one benchmark where Gemini 3 Pro was out-performed by Claude Sonnet 4.5 - Gemini 3 Pro got 76.2% and Sonnet 4.5 got 77.2%. OpenAI now have the highest scoring model there by a full .7 of a percentage point!

They also report a score of 58.1% on Terminal Bench 2.0, beating Gemini 3 Pro's 54.2% (and Sonnet 4.5's 42.8%.)

The most intriguing part of this announcement concerns the model's approach to long context problems:

GPT‑5.1-Codex-Max is built for long-running, detailed work. It’s our first model natively trained to operate across multiple context windows through a process called compaction, coherently working over millions of tokens in a single task. [...]

Compaction enables GPT‑5.1-Codex-Max to complete tasks that would have previously failed due to context-window limits, such as complex refactors and long-running agent loops by pruning its history while preserving the most important context over long horizons. In Codex applications, GPT‑5.1-Codex-Max automatically compacts its session when it approaches its context window limit, giving it a fresh context window. It repeats this process until the task is completed.

There's a lot of confusion on Hacker News about what this actually means. Claude Code already does a version of compaction, automatically summarizing previous turns when the context runs out. Does this just mean that Codex-Max is better at that process?

I had it draw me a couple of pelicans by typing "Generate an SVG of a pelican riding a bicycle" directly into the Codex CLI tool. Here's thinking level medium:

And here's thinking level "xhigh":

I also tried xhigh on the my longer pelican test prompt, which came out like this:

Also today: GPT-5.1 Pro is rolling out today to all Pro users. According to the ChatGPT release notes:

GPT-5.1 Pro is rolling out today for all ChatGPT Pro users and is available in the model picker. GPT-5 Pro will remain available as a legacy model for 90 days before being retired.

That's a pretty fast deprecation cycle for the GPT-5 Pro model that was released just three months ago.

# 19th November 2025, 11:15 pm / llm-release, gpt-5, codex-cli, generative-ai, openai, ai, llms, pelican-riding-a-bicycle, evals, gpt-codex

Introducing GPT-5.1 for developers. OpenAI announced GPT-5.1 yesterday, calling it a smarter, more conversational ChatGPT. Today they've added it to their API.

We actually got four new models today:

• gpt-5.1
• gpt-5.1-chat-latest
• gpt-5.1-codex
• gpt-5.1-codex-mini

There are a lot of details to absorb here.

GPT-5.1 introduces a new reasoning effort called "none" (previous were minimal, low, medium, and high) - and none is the new default.

This makes the model behave like a non-reasoning model for latency-sensitive use cases, with the high intelligence of GPT‑5.1 and added bonus of performant tool-calling. Relative to GPT‑5 with 'minimal' reasoning, GPT‑5.1 with no reasoning is better at parallel tool calling (which itself increases end-to-end task completion speed), coding tasks, following instructions, and using search tools---and supports web search⁠ in our API platform.

When you DO enable thinking you get to benefit from a new feature called "adaptive reasoning":

On straightforward tasks, GPT‑5.1 spends fewer tokens thinking, enabling snappier product experiences and lower token bills. On difficult tasks that require extra thinking, GPT‑5.1 remains persistent, exploring options and checking its work in order to maximize reliability.

Another notable new feature for 5.1 is extended prompt cache retention:

Extended prompt cache retention keeps cached prefixes active for longer, up to a maximum of 24 hours. Extended Prompt Caching works by offloading the key/value tensors to GPU-local storage when memory is full, significantly increasing the storage capacity available for caching.

To enable this set "prompt_cache_retention": "24h" in the API call. Weirdly there's no price increase involved with this at all. I asked about that and OpenAI's Steven Heidel replied:

with 24h prompt caching we move the caches from gpu memory to gpu-local storage. that storage is not free, but we made it free since it moves capacity from a limited resource (GPUs) to a more abundant resource (storage). then we can serve more traffic overall!

The most interesting documentation I've seen so far is in the new 5.1 cookbook, which also includes details of the new shell and apply_patch built-in tools. The apply_patch.py implementation is worth a look, especially if you're interested in the advancing state-of-the-art of file editing tools for LLMs.

I'm still working on integrating the new models into LLM. The Codex models are Responses-API-only.

I got this pelican for GPT-5.1 default (no thinking):

And this one with reasoning effort set to high:

These actually feel like a regression from GPT-5 to me. The bicycles have less spokes!

# 13th November 2025, 11:59 pm / llm, openai, pelican-riding-a-bicycle, llm-reasoning, ai, llms, llm-release, gpt-5, generative-ai, gpt-codex

Kimi K2 Thinking. Chinese AI lab Moonshot's Kimi K2 established itself as one of the largest open weight models - 1 trillion parameters - back in July. They've now released the Thinking version, also a trillion parameters (MoE, 32B active) and also under their custom modified (so not quite open source) MIT license.

Starting with Kimi K2, we built it as a thinking agent that reasons step-by-step while dynamically invoking tools. It sets a new state-of-the-art on Humanity's Last Exam (HLE), BrowseComp, and other benchmarks by dramatically scaling multi-step reasoning depth and maintaining stable tool-use across 200–300 sequential calls. At the same time, K2 Thinking is a native INT4 quantization model with 256k context window, achieving lossless reductions in inference latency and GPU memory usage.

This one is only 594GB on Hugging Face - Kimi K2 was 1.03TB - which I think is due to the new INT4 quantization. This makes the model both cheaper and faster to host.

So far the only people hosting it are Moonshot themselves. I tried it out both via their own API and via the OpenRouter proxy to it, via the llm-moonshot plugin (by NickMystic) and my llm-openrouter plugin respectively.

The buzz around this model so far is very positive. Could this be the first open weight model that's competitive with the latest from OpenAI and Anthropic, especially for long-running agentic tool call sequences?

Moonshot AI's self-reported benchmark scores show K2 Thinking beating the top OpenAI and Anthropic models (GPT-5 and Sonnet 4.5 Thinking) at "Agentic Reasoning" and "Agentic Search" but not quite top for "Coding":

I ran a couple of pelican tests:

llm install llm-moonshot
llm keys set moonshot # paste key
llm -m moonshot/kimi-k2-thinking 'Generate an SVG of a pelican riding a bicycle'

llm install llm-openrouter
llm keys set openrouter # paste key
llm -m openrouter/moonshotai/kimi-k2-thinking \
  'Generate an SVG of a pelican riding a bicycle'

Artificial Analysis said:

Kimi K2 Thinking achieves 93% in 𝜏²-Bench Telecom, an agentic tool use benchmark where the model acts as a customer service agent. This is the highest score we have independently measured. Tool use in long horizon agentic contexts was a strength of Kimi K2 Instruct and it appears this new Thinking variant makes substantial gains

CNBC quoted a source who provided the training price for the model:

The Kimi K2 Thinking model cost $4.6 million to train, according to a source familiar with the matter. [...] CNBC was unable to independently verify the DeepSeek or Kimi figures.

MLX developer Awni Hannun got it working on two 512GB M3 Ultra Mac Studios:

The new 1 Trillion parameter Kimi K2 Thinking model runs well on 2 M3 Ultras in its native format - no loss in quality!

The model was quantization aware trained (qat) at int4.

Here it generated ~3500 tokens at 15 toks/sec using pipeline-parallelism in mlx-lm

Here's the 658GB mlx-community model.

# 6th November 2025, 11:53 pm / llm, llm-reasoning, pelican-riding-a-bicycle, ai, ai-in-china, llms, moonshot, llm-release, generative-ai, openrouter, kimi, artificial-analysis, mlx

Introducing SWE-1.5: Our Fast Agent Model (via) Here's the second fast coding model released by a coding agent IDE in the same day - the first was Composer-1 by Cursor. This time it's Windsurf releasing SWE-1.5:

Today we’re releasing SWE-1.5, the latest in our family of models optimized for software engineering. It is a frontier-size model with hundreds of billions of parameters that achieves near-SOTA coding performance. It also sets a new standard for speed: we partnered with Cerebras to serve it at up to 950 tok/s – 6x faster than Haiku 4.5 and 13x faster than Sonnet 4.5.

Like Composer-1 it's only available via their editor, no separate API yet. Also like Composer-1 they don't appear willing to share details of the "leading open-source base model" they based their new model on.

I asked it to generate an SVG of a pelican riding a bicycle and got this:

This one felt really fast. Partnering with Cerebras for inference is a very smart move.

They share a lot of details about their training process in the post:

SWE-1.5 is trained on our state-of-the-art cluster of thousands of GB200 NVL72 chips. We believe SWE-1.5 may be the first public production model trained on the new GB200 generation. [...]

Our RL rollouts require high-fidelity environments with code execution and even web browsing. To achieve this, we leveraged our VM hypervisor otterlink that  allows us to scale Devin to tens of thousands of concurrent machines (learn more about blockdiff). This enabled us to smoothly support very high concurrency and ensure the training environment is aligned with our Devin production environments.

That's another similarity to Cursor's Composer-1! Cursor talked about how they ran "hundreds of thousands of concurrent sandboxed coding environments in the cloud" in their description of their RL training as well.

This is a notable trend: if you want to build a really great agentic coding tool there's clearly a lot to be said for using reinforcement learning to fine-tune a model against your own custom set of tools using large numbers of sandboxed simulated coding environments as part of that process.

Update: I think it's built on GLM.

# 29th October 2025, 11:59 pm / inference-speed, llm-release, coding-agents, ai-assisted-programming, generative-ai, pelican-riding-a-bicycle, ai, llms, glm

MiniMax M2 & Agent: Ingenious in Simplicity. MiniMax M2 was released on Monday 27th October by MiniMax, a Chinese AI lab founded in December 2021.

It's a very promising model. Their self-reported benchmark scores show it as comparable to Claude Sonnet 4, and Artificial Analysis are ranking it as the best currently available open weight model according to their intelligence score:

MiniMax’s M2 achieves a new all-time-high Intelligence Index score for an open weights model and offers impressive efficiency with only 10B active parameters (200B total). [...]

The model’s strengths include tool use and instruction following (as shown by Tau2 Bench and IFBench). As such, while M2 likely excels at agentic use cases it may underperform other open weights leaders such as DeepSeek V3.2 and Qwen3 235B at some generalist tasks. This is in line with a number of recent open weights model releases from Chinese AI labs which focus on agentic capabilities, likely pointing to a heavy post-training emphasis on RL.

The size is particularly significant: the model weights are 230GB on Hugging Face, significantly smaller than other high performing open weight models. That's small enough to run on a 256GB Mac Studio, and the MLX community have that working already.

MiniMax offer their own API, and recommend using their Anthropic-compatible endpoint and the official Anthropic SDKs to access it. MiniMax Head of Engineering Skyler Miao provided some background on that:

M2 is a agentic thinking model, it do interleaved thinking like sonnet 4.5, which means every response will contain its thought content. Its very important for M2 to keep the chain of thought. So we must make sure the history thought passed back to the model. Anthropic API support it for sure, as sonnet needs it as well. OpenAI only support it in their new Response API, no support for in ChatCompletion.

MiniMax are offering the new model via their API for free until November 7th, after which the cost will be $0.30/million input tokens and $1.20/million output tokens - similar in price to Gemini 2.5 Flash and GPT-5 Mini, see price comparison here on my llm-prices.com site.

I released a new plugin for LLM called llm-minimax providing support for M2 via the MiniMax API:

llm install llm-minimax
llm keys set minimax
# Paste key here
llm -m m2 -o max_tokens 10000 "Generate an SVG of a pelican riding a bicycle"

Here's the result:

51 input, 4,017 output. At $0.30/m input and $1.20/m output that pelican would cost 0.4836 cents - less than half a cent.

This is the first plugin I've written for an Anthropic-API-compatible model. I released llm-anthropic 0.21 first adding the ability to customize the base_url parameter when using that model class. This meant the new plugin was less than 30 lines of Python.

# 29th October 2025, 10:49 pm / llm-release, generative-ai, pelican-riding-a-bicycle, llm-pricing, ai, ai-in-china, llms, local-llms, llm, minimax

Composer: Building a fast frontier model with RL (via) Cursor released Cursor 2.0 today, with a refreshed UI focused on agentic coding (and running agents in parallel) and a new model that's unique to Cursor called Composer 1.

As far as I can tell there's no way to call the model directly via an API, so I fired up "Ask" mode in Cursor's chat side panel and asked it to "Generate an SVG of a pelican riding a bicycle":

Here's the result:

The notable thing about Composer-1 is that it is designed to be fast. The pelican certainly came back quickly, and in their announcement they describe it as being "4x faster than similarly intelligent models".

It's interesting to see Cursor investing resources in training their own code-specific model - similar to GPT-5-Codex or Qwen3-Coder. From their post:

Composer is a mixture-of-experts (MoE) language model supporting long-context generation and understanding. It is specialized for software engineering through reinforcement learning (RL) in a diverse range of development environments. [...]

Efficient training of large MoE models requires significant investment into building infrastructure and systems research. We built custom training infrastructure leveraging PyTorch and Ray to power asynchronous reinforcement learning at scale. We natively train our models at low precision by combining our MXFP8 MoE kernels with expert parallelism and hybrid sharded data parallelism, allowing us to scale training to thousands of NVIDIA GPUs with minimal communication cost. [...]

During RL, we want our model to be able to call any tool in the Cursor Agent harness. These tools allow editing code, using semantic search, grepping strings, and running terminal commands. At our scale, teaching the model to effectively call these tools requires running hundreds of thousands of concurrent sandboxed coding environments in the cloud.

One detail that's notably absent from their description: did they train the model from scratch, or did they start with an existing open-weights model such as something from Qwen or GLM?

Cursor researcher Sasha Rush has been answering questions on Hacker News, but has so far been evasive in answering questions about the base model. When directly asked "is Composer a fine tune of an existing open source base model?" they replied:

Our primary focus is on RL post-training. We think that is the best way to get the model to be a strong interactive agent.

Sasha did confirm that rumors of an earlier Cursor preview model, Cheetah, being based on a model by xAI's Grok were "Straight up untrue."

# 29th October 2025, 8:45 pm / inference-speed, llm-release, cursor, pelican-riding-a-bicycle, generative-ai, ai, llms, parallel-agents, coding-agents, ai-assisted-programming

Introducing Claude Haiku 4.5 (via) Anthropic released Claude Haiku 4.5 today, the cheapest member of the Claude 4.5 family that started with Sonnet 4.5 a couple of weeks ago.

It's priced at $1/million input tokens and $5/million output tokens, slightly more expensive than Haiku 3.5 ($0.80/$4) and a lot more expensive than the original Claude 3 Haiku ($0.25/$1.25), both of which remain available at those prices.

It's a third of the price of Sonnet 4 and Sonnet 4.5 (both $3/$15) which is notable because Anthropic's benchmarks put it in a similar space to that older Sonnet 4 model. As they put it:

What was recently at the frontier is now cheaper and faster. Five months ago, Claude Sonnet 4 was a state-of-the-art model. Today, Claude Haiku 4.5 gives you similar levels of coding performance but at one-third the cost and more than twice the speed.

I've been hoping to see Anthropic release a fast, inexpensive model that's price competitive with the cheapest models from OpenAI and Gemini, currently $0.05/$0.40 (GPT-5-Nano) and $0.075/$0.30 (Gemini 2.0 Flash Lite). Haiku 4.5 certainly isn't that, it looks like they're continuing to focus squarely on the "great at code" part of the market.

The new Haiku is the first Haiku model to support reasoning. It sports a 200,000 token context window, 64,000 maximum output (up from just 8,192 for Haiku 3.5) and a "reliable knowledge cutoff" of February 2025, one month later than the January 2025 date for Sonnet 4 and 4.5 and Opus 4 and 4.1.

Something that caught my eye in the accompanying system card was this note about context length:

For Claude Haiku 4.5, we trained the model to be explicitly context-aware, with precise information about how much context-window has been used. This has two effects: the model learns when and how to wrap up its answer when the limit is approaching, and the model learns to continue reasoning more persistently when the limit is further away. We found this intervention—along with others—to be effective at limiting agentic “laziness” (the phenomenon where models stop working on a problem prematurely, give incomplete answers, or cut corners on tasks).

I've added the new price to llm-prices.com, released llm-anthropic 0.20 with the new model and updated my Haiku-from-your-webcam demo (source) to use Haiku 4.5 as well.

Here's llm -m claude-haiku-4.5 'Generate an SVG of a pelican riding a bicycle' (transcript).

18 input tokens and 1513 output tokens = 0.7583 cents.

# 15th October 2025, 7:36 pm / llm, anthropic, claude, llm-reasoning, llm-pricing, ai, llms, llm-release, generative-ai, pelican-riding-a-bicycle

GPT-5 pro. Here's OpenAI's model documentation for their GPT-5 pro model, released to their API today at their DevDay event.

It has similar base characteristics to GPT-5: both share a September 30, 2024 knowledge cutoff and 400,000 context limit.

GPT-5 pro has maximum output tokens 272,000 max, an increase from 128,000 for GPT-5.

As our most advanced reasoning model, GPT-5 pro defaults to (and only supports) reasoning.effort: high

It's only available via OpenAI's Responses API. My LLM tool doesn't support that in core yet, but the llm-openai-plugin plugin does. I released llm-openai-plugin 0.7 adding support for the new model, then ran this:

llm install -U llm-openai-plugin
llm -m openai/gpt-5-pro "Generate an SVG of a pelican riding a bicycle"

It's very, very slow. The model took 6 minutes 8 seconds to respond and charged me for 16 input and 9,205 output tokens. At $15/million input and $120/million output this pelican cost me $1.10!

Here's the full transcript. It looks visually pretty simpler to the much, much cheaper result I got from GPT-5.

# 6th October 2025, 7:48 pm / llm-release, generative-ai, openai, pelican-riding-a-bicycle, ai, llms, gpt-5, llm-reasoning, llm-pricing