← All updates

Mistral Opens a Trillion-Parameter Model

Plus: Pika Releases H3 Flex, Claude API Raises Request Limits

Good Morning!

Mistral open-weights a trillion-parameter model while our ranking's top stays cheap Flash-tier builds: competitive pricing could help a new open release win adoption.

GPT-6 Luna climbs to second on Oct 29, shoving GLM 5.3 Flash down a seat. Claude Sonnet 5.5 takes the tenth spot. Luna's own traffic nearly doubled day over day, to 2.76T, and general work is its largest lane.

DIT
Model Ranking
DAILY
Top 10 text models by real-world token usage · Oct 29, 2026 (UTC)
Real-world use:
Agent
Code
General
Data
Model
tokens · vs. previous day (Oct 28)
1
DeepSeek V4.1 Flash
· DeepSeek
10.28T
–
+6%
2
GPT-6 Luna
· OpenAI
2.76T
▲1
+88%
3
GLM 5.3 Flash
· Z.ai
2.52T
▼1
+4%
4
Hy4 preview
· Tencent
1.68T
–
+16%
5
Claude Opus 5.5
· Anthropic
1.31T
▲1
+30%
6
GPT-6.1 Sol
· OpenAI
1.13T
▲3
+113%
7
MiniMax M3.1 Flash
· MiniMax
0.97T
–
+7%
8
DeepSeek V4 Flash
· DeepSeek
0.93T
▼3
-9%
9
GLM 5.3
· Z.ai
0.66T
▼1
+4%
10
Claude Sonnet 5.5
· Anthropic
0.57T
▲2
+27%
How we rank: tokens processed over the full 24 hours of Oct 29, 2026 (00:00–24:00 UTC), counted and classified by DIT Radar's internal model.
Excluded: StepFun Step 5 Preview, whose traffic showed an anomalous, non-organic pattern.
Source: OpenRouter. The ranking reflects DIT Radar's interpretation of the data.

Mistral Open-Weights Trillion-Parameter Le Chonk

Photo: Mistral

TL;DR: Mistral released the weights for Le Chonk, a trillion-parameter model trained entirely in Europe. The lab reports a lead among open-weight models on Cognition's FrontierCode 2.0 coding benchmark and a separate 51% score on Artificial Analysis' Cyber Index. Independent testing finds fast API output. For teams that want self-hosted coding agents, a top-tier open build is now a real option. Read more →


Pika Open-Weights Video Model H3 Flex

Photo: André Eusébio, Unsplash

TL;DR: Pika released H3 Flex, an open-weight text-to-video model it says is the top open model on DesignArena's board. The open release supports native 480p and 768p video with synchronized audio. Up to 4K upscaling, audio-stem separation and the accelerated serving stack are Pika-hosted features. Its faster-than-real-time claim measures model computation under benchmark conditions, not guaranteed end-to-end latency. Read more →


Anthropic Lifts Claude API Request Size

Photo: Brecht Corbeel, Unsplash

TL;DR: Anthropic raised the maximum request size for the Messages and Token Counting endpoints on its first-party Claude API from 32 MB to 128 MB, allowing larger payloads such as PDFs and image-heavy prompts. Context-window and PDF page limits still apply. A practical limit change for anyone pushing big documents through the API. Read more →


Google Releases Gemini 4 Carbon Frontier Model

Photo: Albert Stoynov, Unsplash

TL;DR: Google released Gemini 4 Carbon, its newest frontier model, claiming state-of-the-art results in cybersecurity and long-horizon tasks. Access starts with trusted cyber defenders, so wider API availability is still limited. Google has announced introductory pricing of $2 per million input tokens and $10 per million output tokens; wider access remains pending. Read more →


Mistral Le Chonk Leads Europe, Trails China

Image: The Decoder

TL;DR: Independent testing finds fast API output from Mistral's open-weight Le Chonk, but it still trails leading Chinese models overall. Mistral reports leads among open-weight models on FrontierCode 2.0 and Terminal-Bench 4.0, while matching GLM-5.3 on Artificial Analysis' Coding Agent Index. A strong European option for coding workloads, not yet a frontier general model. Read more →


Gemini 4 Carbon Tops Independent Evaluations

Photo: Conny Schneider, Unsplash

TL;DR: Google's new frontier model, Gemini 4 Carbon, tops Artificial Analysis' Intelligence Index v5 with a score of 59 and launches first for trusted cyber defenders. The staged rollout keeps general access unclear, so teams planning around it should track when the public API opens. Availability, not benchmarks, is the constraint for now. Read more →


Nanbeige Open-Weights 4.3-33B Teacher Model

Image: Hugging Face

TL;DR: Nanbeige published Nanbeige4.3-33B, an open-weight model positioned as a flagship teacher for distillation. The weights and model card are public on Hugging Face, giving builders another mid-size base to fine-tune and self-host. It is a smaller-lab release, so check its benchmarks before adopting. Read more →


Top AI models. Lower costs.

Access GPT, Claude, Gemini, and more through one OpenAI-compatible API—and save up to 70% on AI API costs. Create your DIT account, get your API key, and start building.