Breaking
Advertisement

AI

AI

Mirendil Launches Fern 0.1, Its First Model, With 10M-Token Context

Mirendil has released Fern, its first AI model. Mirendil Fern is a from-scratch coding model with a context window of up to 10M tokens. Mirendil explains how that equals about 23 MB of code, or roughly half a million lines. That is a codebase on the scale of Megatron-LM, held in context at once. Fern 0.1 is the first version, built from Mirendil-Base after a short round of post-training. For developers, its like, you load a whole repo, a full training log or days of agent traces at once. TL;DR Size: 0.1 to 0.5T parameters (exact count and active params not disclosed); context up to 10M tokens. Runs on: Hosted early access only; no open weights or quantized builds announced. Performance: Strong long-context retrieval, mid-pack agentic coding versus open-weight models, well behind frontier models. Best: 100% on RULER single-needle retrieval at every length from 32K to 10M tokens (base model). Worst: 6.1% on Terminal Bench 4.0, versus 66.4% for Claude Opus 5.5 in Mirendil’s own table. What is Mirendil Fern? Fern is a language model from Mirendil, an AI R&D lab. Fern is aimed at AI R&D work, not general chat. Mirendil lists 6 target uses: Working across larger codebases Agentic coding that runs commands until tests pass Research code: training loops, data pipelines and evaluations Debugging training runs from long logs and stack traces Reading full agent or training traces end to end Auditing pretraining shards for duplicates, contamination and junk How was Fern built? Fern sits on Mirendil-Base, which Mirendil says one researcher pretrained in 3.5 months. Total compute was under $10M, including experiments, agent token spend and data processing. AI agents wrote over 99% of the code. The researcher chose what to try and when to intervene. Training data came only from public sources: GitHub, public code, web pages and public datasets. Mirendil says it used no distillation, no synthetic data and no licensed data. The code was written from scratch on PyTorch and JAX. Fern is the smallest of 3 planned models. Mirendil-Base-Large (0.5 to 2T parameters) is over 80% through pretraining. Mirendil-Base-XL (2 to 10T) is over 20% through. Does longer context actually help? Mirendil tested next-token prediction on its own repositories, concatenated into a 42.9 MB stream. It measured cross-entropy in bits per byte, where lower is better. Window sizes here are measured in bytes, not tokens. Fern’s loss fell from 0.246 at a 32K-byte window to 0.155 at 16M and 23M bytes. The other models were scored up to about 1M tokens, their advertised limit. At a 1M-byte window, Kimi-K3 scored 0.140 and Fern 0.165. So Fern is not the best predictor at equal context. Its advantage is that the curve keeps going past where others stop. How does Fern score on RULER? RULER is a standard long-context benchmark. Mirendil ran 5 RULER tasks on Mirendil-Base, the pre-post-training model. It used greedy decoding, thinking off and exact string matching. Retrieval holds up well. Multi-hop reasoning drops to 45% at 10M. Each 10M cell uses only 20 examples, versus 100 elsewhere. Mirendil says broader long-context results are coming. (function(){var f=document.getElementById('mtp-fern-frame');window.addEventListener('message',function(e){if(e.source===f.contentWindow&&e.data&&e.data.mtpFernH){f.style.height=e.data.mtpFernH+'px'}});})(); How does Fern 0.1 do on agentic benchmarks? Mirendil compared Fern 0.1 with open-weight and frontier models. Scores for other models come from model cards, leaderboards or Mirendil’s own runs. Fern 0.1 beats Inkling and Nemotron 3 Ultra on 4 of 5 tests. It ties Beam on SWE-Pro v2 Hard at 76.5. It trails Beam on SWE-Bench Verified, 73.4 versus 81.6. Every frontier and large open model in the table leads it. How does Fern compare with similar models? *Figures from Mirendil’s announcement charts. Key Takeaways Fern 0.1 supports up to 10M tokens, about 23 MB of code. Bits per byte keeps falling to 0.155 at a 23M-byte window. RULER single needle stays at 100% from 32K to 10M tokens. Agentic coding lags: 6.1% on Terminal Bench 4.0. Base model built by 1 researcher plus agents for under $10M. Check out the Fern announcement, the Mirendil-Base pretraining post, and request early access. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well. Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us The post Mirendil Launches Fern 0.1, Its First Model, With 10M-Token Context appeared first on MarkTechPost.

AI

OpenAI asks outside experts, including a Nobel laureate, to check proofs its AI produced in math, computer science, and physics

OpenAI says its internal AI models have cracked three major open questions in math, computer science, and physics. One of them is part of a famous Millennium Prize Problem, each of which carries a $1 million prize. Outside experts, including a Nobel laureate, are now checking whether the proofs hold up. If the computer science proof holds on its own, hundreds of research papers lose their foundation. The article OpenAI asks outside experts, including a Nobel laureate, to check proofs its AI produced in math, computer science, and physics appeared first on The Decoder.

Advertisement
AI

Anthropic uses its IPO letter to renew its pledge that Claude will never show ads

Anthropic says it will keep ads out of Claude. In a cover letter accompanying its IPO filing, the company argues that the chatbot's sole incentive should be to provide helpful answers. The article Anthropic uses its IPO letter to renew its pledge that Claude will never show ads appeared first on The Decoder.

AI

OpenAI’s Ted Sanders shares a rule of thumb for picking the right GPT-model

OpenAI's Ted Sanders explains where GPT-6.1 Terra fits in the company's GPT lineup. The smaller model outperforms the larger Astra in technical communication and costs far less but trails it in most other areas. Sanders says model size, reasoning effort, and multi-agent systems are distinct dimensions that can’t be reduced to a single "IQ score." The article OpenAI's Ted Sanders shares a rule of thumb for picking the right GPT-model appeared first on The Decoder.

Advertisement
AI

vLLM Semantic Router Releases d3-ultra: A 27B Open Multimodal Decision Model

The vLLM Semantic Router team has released d3-ultra, a 27B multimodal decision model under Apache-2.0. The d3-ultra decision model does not generate text. It reads an input once and returns a probability for every answer to every question. For developers, that means routing, classification, visual grounding and video checks from 1 open-weight call. TL;DR Size: 26.09B parameters (dense, including a 0.46B vision encoder); 131,072 tokens of context per question. Runs on: tested on 1 AMD Instinct MI325X with BF16 weights; no official quantized variant is listed yet. Performance: highest SR-Text score of the 6 decision models in the team’s own private test. Best: 97.8 on SR-Text tools and automation (chance-corrected accuracy). Worst: 54.6% on SR-Box (9 domains), 5.0 points below its sibling d3-turbo. What is d3-ultra? d3-ultra is a System One decision model from vLLM Semantic Router that answers typed questions with probabilities. It is the newest member of Decision 3.0, a family that spans 0.6B to 27B. You send a state (text or JSON, plus images or videos) and a set of questions. Each question has a type: Choice: pick 1 of several options. Yes / No: return the probability of yes. Score: rate on a scale. Box: locate something in an image or a video moment. Track: follow a target across a whole video. Images can also sit inside questions and options. A quiz can show a photo, then offer 4 etchings as answers. How does d3-ultra work? The core idea is state once. The model reads the input once per request and answers every question from that single read. The config shows a Qwen3.5-architecture backbone with 64 layers. 48 are linear-attention layers and 16 are full-attention layers. The card does not name its base checkpoint. vLLM’s recipe says the sibling d3 was trained from Qwen3.8-27B. Box answers always return exactly 1 box with a confidence. A confidence near 0 means the target is not in the picture. Questions over 131,072 tokens fail with max_length_exceeded; nothing is silently truncated. How does d3-ultra perform on benchmarks? The research team compares d3-ultra with GLiDE V1.2, d1-ultra, d3-turbo, d3 and Jev 1.14 on 2 in-house tests. SR-Text has 10,411 text requests with 73,495 questions. Most recreate skills from 34 public benchmarks with new items. Another 4,896 questions come from 43 new task families, like SQL checks and ticket routing. Scores are chance-corrected: 0 is chance, 100 is perfect. d3-ultra: 87.2 GLiDE V1.2: 85.1 d1-ultra: 83.4 d3-turbo: 83.1 d3: 81.4 Jev 1.14: 73.1 d3-ultra leads every area except arts and human taste. There, d1-ultra scores 90.8 against 89.6, within the margin of error. SR-Box has 5,678 Box questions on 4,984 images across 10 domains. A box counts as correct at IoU 0.5 or higher. d3-ultra scores 58.2% over 10 domains and 54.6% over 9. d3-turbo scores 60.0% and 59.6%. Jev 1.14 scores 45.7% and 46.9%. The team says d3-turbo stays stronger on natural images, while d3-ultra leads on charts. Latency: on 1 AMD Instinct MI325X with AutoModel, 1 question about a photo takes 304 ms. 255 questions take 2.16 s, about 7.3 ms per extra question. (function(){var f=document.getElementById('mtp-d3u-frame'); window.addEventListener('message',function(e){if(e.data&&e.data.mtpD3u&&e.source===f.contentWindow){f.style.height=e.data.height+'px';}});})(); How do you deploy d3-ultra? There are 4 routes: Transformers: AutoModel.from_pretrained("vllm-sr/d3-ultra", trust_remote_code=True) with transformers==5.17.0. The optional flash-linear-attention package adds fast GPU kernels. vLLM Semantic Router Engine: vllm-sr serve -e vllm-sr/d3-ultra, added in PR #5449 on October 25, 2026. Free API: the card says decision.vllm-sr.ai serves d3-ultra, with a free key. Demo: the official Hugging Face Space. d3-ultra vs GLiDE V1.2, d1-ultra, Jev 1.14 and d3-turbo Key Takeaways d3-ultra is a 27B Apache-2.0 decision model that never generates text. It reads input once: 255 questions about 1 photo take 2.16 s. It tops the team’s SR-Text test at 87.2, ahead of GLiDE V1.2 at 85.1. It trails d3-turbo on Box grounding: 58.2% vs 60.0%. All benchmark numbers are self-reported on private test sets. Check out the model card, the GitHub repo, the Decision 3.0 collection and the Hugging Face Space. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well. Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us The post vLLM Semantic Router Releases d3-ultra: A 27B Open Multimodal Decision Model appeared first on MarkTechPost.

AI

Baidu Open-Sources Unlimited OCR 2 for Hundred-Page Document Parsing

Baidu has released Unlimited OCR 2, an open-source OCR model built for very long documents. The 3B model parses books, filings and reports of hundreds of pages in one request. Its core change is SR-SWA, a sparse attention scheme that keeps per-step decoding cost nearly flat. Weights ship under the MIT license, with code on GitHub. TL;DR Size: 3.34B parameters. Active parameters not disclosed. Max position length 131,072 tokens. Runs on: 1 CUDA GPU in BF16. Baidu tested on an NVIDIA A800-80GB. BF16 weights total about 6.7 GB. Performance: first place on both multi-page benchmarks Baidu tested, against 9 other models (self-reported). Best: 77.1 overall on MarathonDocBench (hundreds of pages). The next model, Kimi-K3, scores 33.7. Worst: 48.2 on heading hierarchy (TEDS) on MarathonDocBench, its lowest sub-score. What is Unlimited OCR 2? Unlimited OCR 2 is a 3B vision-language model from Baidu that turns ordered page images into structured text. Output covers text, tables, formulas, figures, headings and reading order. It succeeds Unlimited-OCR, released in June 2026 to push DeepSeek-OCR further. The config keeps a DeepSeek-OCR style layout. A SAM ViT-B plus CLIP-L encoder feeds a 12-layer mixture-of-experts decoder. Each token routes to 6 of 64 experts, plus 2 shared experts. The DeepSeek-OCR paper describes this decoder as DeepSeek3B-MoE-A570M. Baidu does not state an active count for version 2. In multi-page mode, each page costs 273 image tokens. How does SR-SWA work? Full attention cost grows quadratically with sequence length. During decoding, each new token scans a longer KV cache. At 273 image tokens per page, 100 pages add 27,300 input tokens before any output. SR-SWA stands for Sparse Reference Sliding Window Attention. It compresses text-visual attention into a fixed set of working references. Baidu’s model card gives 512 tokens as the example size. The config pairs a 128-token local window with top-16 selection over 32-token blocks. That is 16 blocks of 32 tokens, matching the 512-token budget. This keeps per-step latency and memory nearly constant as documents grow. It calls the result ‘unlimited-horizon’ parsing within a finite context window. Image encoding and total runtime still rise with page count. #mtp-uo2-embed{background:transparent!important;border:0!important;} #mtp-uo2-embed iframe{width:100%!important;border:0!important;display:block!important;background:transparent!important;} #mtp-uo2-embed hr,#mtp-uo2-embed p:empty,#mtp-uo2-embed del,#mtp-uo2-embed s{display:none!important;} (function(){var f=document.getElementById('mtp-uo2-frame');window.addEventListener('message',function(e){if(e.data&&e.data.mtpEmbed==='uo2'&&e.data.height&&e.source===f.contentWindow){f.style.height=e.data.height+'px';}});})(); How does it score on MPDocBench and MarathonDocBench? Baidu team tests on MPDocBench, with dozens of pages per document, and MarathonDocBench, with hundreds. MPDocBench-Parse is a public benchmark of 433 documents and 3,246 pages. We could not find a public MarathonDocBench release. Per Baidu’s leaderboard chart, Unlimited OCR 2 scores 84.6 on MPDocBench. Qwen3.8-Max follows at 78.3, then Qwen-3.8-Flash-Next at 77.4. On MarathonDocBench it scores 77.1. Kimi-K3 is next at 33.7, then Gemini-3.8-Flash at 28.6. The drop between benchmarks is the real story. Qwen3.8-Max falls from 78.3 to 22.9 on hundred-page documents. Unlimited OCR 2 falls 7.5 points. Its predecessor, Unlimited-OCR, scores 0.0 on MarathonDocBench. Sub-scores (MPDocBench / MarathonDocBench): Text: 94.8 / 89.6 Formula: 89.4 / 87.9 Reading order: 92.3 / 85.3 Tables: 87.8 / 74.8 Heading hierarchy: 60.1 / 48.2 Single-metric margins on MPDocBench are thin. Qwen3.8-Max scores 94.3 on text and 89.2 on formulas. All figures are self-reported and appear only in a chart image. Comparison with other open OCR models Parameter counts and licenses from each model’s Hugging Face page. Benchmark scores from Baidu’s chart. How do you run Unlimited OCR 2? The model card lists Transformers for quick tests and SGLang for serving. In Transformers, infer handles 1 page with dynamic tiling, called gundam mode. infer_multi parses every page of a document in 1 request, so the model keeps cross-page context. Loading requires trust_remote_code=True, which runs Python code from the repo. SGLang serving uses a customized wheel from Baidu, not upstream SGLang. Baidu’s example schedules 10 pages per window and reuses the previous window’s KV tail. Smaller windows cut peak memory but weaken cross-page merging. Baidu warns a default bias setting can trigger a 462 MiB allocation and run out of memory. Key Takeaways Unlimited OCR 2 is a 3.34B, MIT-licensed OCR model from Baidu. SR-SWA caps attention at a fixed 512-token reference set. It scores 77.1 on MarathonDocBench, against 33.7 for the next model. Heading hierarchy is the weak spot, at 48.2 on long documents. All benchmark numbers are Baidu’s own. Check out the model card, the technical report and the GitHub repo. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well. Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us The post Baidu Open-Sources Unlimited OCR 2 for Hundred-Page Document Parsing appeared first on MarkTechPost.

AI

OpenAI Launches GPT-6.1 Terra, Halving GPT-6.1 Sol’s API Price for Agent Work

OpenAI has released GPT-6.1 Terra, a fast model for visual reasoning, computer use and writing. It replaces GPT-5.6 Terra and is live in the API as gpt-6.1-terra. Standard pricing is $1 per million input tokens and $5 per million output tokens. For teams automating screens, forms and documents at volume, GPT-6.1 Terra changes the cost math. TL;DR Size: 1.05M-token context, 128K max output. Runs on: OpenAI-hosted only (API, ChatGPT Work, Codex). Closed weights. Performance: Near GPT-6.1 Sol and GPT-6 Astra on vision and computer use, at lower cost. Best: 85.5% on MMMU-Pro (Vision) at max effort, 1.2 points behind GPT-6 Astra. Worst: 50.3% on Terminal-Bench 4.0, behind GPT-6.1 Sol (55.9%) and Claude Sonnet 5.5 (70.6%). What is GPT-6.1 Terra? GPT-6.1 Terra is an OpenAI reasoning model built for high-volume visual and computer-use work. It accepts text and image input and returns text. The context window is 1,050,000 tokens, with up to 128,000 output tokens. Its knowledge cutoff is June 1, 2026. Reasoning effort runs from none to max, with medium as the default. Tool calling requires the Responses API. Supported tools include computer use, hosted shell, code interpreter, web search, file search and MCP. OpenAI’s listed use cases include matching product photos to catalog entries and interpreting technical drawings. How does GPT-6.1 Terra perform on computer use? OpenAI’s main influential test is Agents’ Last Exam v1, a set of professional computer tasks. One task turns architectural drawings into a 3D model. At medium effort, Terra scores 57.6% at $0.29 per task. GPT-5.6 Terra scored 42.6% at $0.84 on the same setting. GPT-6 Astra also scores 57.6% at medium, but costs $4.82 per task. At max effort, Terra reaches 60.0% at $1.01 per task. Claude Opus 5.5 scores 63.2% at $8.82 per task on the same chart. OpenAI says Terra completes tasks over 3 times faster than GPT-6 Astra and GPT-6.1 Sol. On the OSWorld 2.1 offline set, Terra scores 66.7% at medium and 74.2% at max. GPT-5.6 Terra peaked at 57.1%. GPT-6.1 Sol reaches 78.0%, GPT-6 Astra 80.9% and Claude Sonnet 5.5 83.9%. How strong is its visual reasoning? On MMMU-Pro (Vision), Terra scores 85.5% at max effort. GPT-6.1 Sol scores 86.3% and GPT-6 Astra 86.7%. GPT-5.6 Terra topped out at 81.7%. Terra’s max-effort run costs $0.012 per task, against $0.065 for Astra. window.addEventListener("message",function(e){if(e.data&&e.data.mtpTerraH){var f=document.getElementById("mtp-t61-frame");if(f){f.style.height=e.data.mtpTerraH+"px";}}}); Did factuality and coding improve? On OpenAI’s internal set of error-inducing prompts, the low-effort error rate fell from 53.6% to 16.7%. At max effort, it fell 18 points to 4.3%. The set is adversarial by design, so production error rates will differ. On Terminal-Bench 4.0, Terra solves 50.3% at max effort, up from 21.5% for GPT-5.6 Terra. It still trails GPT-6.1 Sol (55.9%), GPT-6 Astra (58.2%) and Claude Sonnet 5.5 (70.6%). OpenAI itself recommends Sol or Astra for the most complex coding tasks. On Terminal-Bench-Science 0.1, Terra scores 58.3% at $4.15 per task. That beats GPT-6.1 Sol’s 57.0% at $5.47. Claude Opus 5.5 with fallbacks scores 63.3% at $23.21. How much does GPT-6.1 Terra cost, and where is it available? Standard API pricing is $1 input, $0.05 cached input and $5 output per million tokens. Batch and Flex are 50% cheaper. Fast mode costs 2x Standard. Prompts above 272K input tokens bill at 2x input and 1.5x output. That is half of GPT-6.1 Sol’s $2/$10 and one-tenth of GPT-6 Astra’s $10/$50. Claude Opus 5.5 lists at $4/$20. In ChatGPT, Terra is live for Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex. GPT-6.1 Terra vs competitors Benchmark scores and per-task costs are read from OpenAI’s launch charts and are self-reported. Specs and prices come from each vendor’s model page. Claude prices: Anthropic pricing. Key Takeaways GPT-6.1 Terra costs $1/$5 per million tokens, half of GPT-6.1 Sol. It matches GPT-6 Astra’s 57.6% on Agents’ Last Exam at $0.29 per task. MMMU-Pro hits 85.5%, within 1.2 points of Astra. Coding still lags: 50.3% on Terminal-Bench 4.0 vs Sonnet 5.5’s 70.6%. All scores are OpenAI-reported; parameters and hardware are undisclosed. Check out the OpenAI announcement, the API model page and the system card addendum. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well. Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us The post OpenAI Launches GPT-6.1 Terra, Halving GPT-6.1 Sol’s API Price for Agent Work appeared first on MarkTechPost.

Advertisement
AI

GPT-6.1 Terra halves OpenAI’s API prices and still gains ground on its larger models

OpenAI has released GPT-6.1 Terra, a mid-range model that cuts API prices in half compared to its predecessor. Analysts say it sets a new standard for intelligence per dollar, especially for everyday computing tasks and AI-driven computer control. The article GPT-6.1 Terra halves OpenAI's API prices and still gains ground on its larger models appeared first on The Decoder.