News
What's happening in robotics and AI — curated by the wujec.ai editors.
DeepSeek's new model beats Opus and GPT on the agent test everyone has already saturated. On the two harder versions of that same test, it loses by twenty points
DeepSeek published V4.1-Flash on 10 September 2026 with a comparison table that reads like a clean sweep — until you follow one test through three of its versions. On Terminal-Bench 2.1 the new model scores 90.6, ahead of Opus-5.0 at 89.1 and GPT-5.6 Sol at 88.8. On Terminal-Bench 3.0 it scores 30.0 against Opus-5.0's 43.3. On version 4.0, 31.2 against 51.8. The three numbers describe the same model on the same family of tasks, and the spread is a lesson in how to read a benchmark. Version 2.1 has been in circulation long enough for every serious laboratory to score in the high eighties; a 1.5-point lead there separates models that are, in practice, equally capable. Versions 3.0 and 4.0 are the harder revisions written precisely because the old one stopped telling models apart — and there the distance between DeepSeek's model and the most expensive Western one is roughly twenty points. The same pattern repeats elsewhere in the table. On Humanity's Last Exam, a test still far from saturated, V4.1-Flash reaches 36.8 against Opus-5.0's 56.3. On ProgramBench, 20.3 against 37.0. What the model does win is worth stating plainly, because it is not a small thing. Its Codeforces rating of 3,471 is the highest in the table. It resolves 74.2 percent of DeepSWE v1.1 tasks, leads on CyberGym with 88.1 and on AutomationBench with 54.8. And it does this while activating 8 billion parameters per token when reading a prompt and 16 billion when writing — out of 552 billion in the backbone. The efficiency is the actual headline. DeepSeek rebuilt the part of the model that stores the conversation, cutting the memory held per token to 890 bytes, about a quarter of what the July model needed. Through the company's API the new model costs $0.30 per million input tokens at peak and $1.20 per million output, against $0.44 and $1.32 for its predecessor, with off-peak hours at half price. Weights are MIT-licensed and were downloaded 140,636 times in the following weeks. Read together, the table says something more useful than 'best model'. For agent work at scale, where a million-token session runs on a budget, this is a remarkable instrument at a fraction of frontier pricing. For the hardest reasoning a buyer can currently pose, the expensive models remain ahead, and DeepSeek's own numbers say so.
DeepSeek-V4.1-Flash →XPeng's robot unit is valued at $6.3bn before it has a single customer
XPeng announced on 24 August 2026 that its humanoid robotics business had raised more than US$900 million at a post-money valuation of over US$6.3 billion. By the company's own timeline, that valuation was set roughly a year and a half before the first robot is due to reach a paying customer. The round was led by IDG Capital, with Gaorong Ventures participating and Tencent and Alibaba coming in as strategic investors. XPeng calls it the largest single private financing round ever recorded in China's embodied AI industry. The carmaker keeps control: the robotics business stays a consolidated subsidiary. The schedule attached to the money is the part worth reading twice. Mass production of the IRON humanoid is expected at the end of 2026. The first units go to XPeng's own stores and campuses. Official launch and customer deliveries, in China and abroad, are scheduled for 2027. That sequence means the $6.3 billion is not a multiple of anything. When Unitree listed in Shanghai five days earlier, its closing price worked out to 201 times last year's sales — an extreme number, but a number, because there were sales to divide by. Here there is no denominator. The first party to take delivery of an IRON is XPeng itself. What the money buys is a machine the company describes as 76 degrees of freedom across the body and 21 in each hand, wrapped in a proprietary fully enclosed flexible lattice structure and driven by three in-house Turing AI chips rated at a combined 2,250 TOPS. Those chips are the strongest argument for the valuation: they are the same silicon XPeng already builds for its cars, which makes the robot programme an extension of an existing supply chain rather than a start from zero. One caveat on the specifications. XPeng's Chinese product page for IRON still states 22 degrees of freedom per hand, while the English-language release issued this week says 21. Both are first-party documents published in the same week. We keep the figure from the product page in our profile and will update it when the company reconciles the two. No price for IRON has been published, no order book has been disclosed, and no independent party has tested the robot.
XPeng IRON →Unitree's whole deployment line has no legs — all three D-series humanoids stand on a column
Unitree lists five humanoids under "Application Plan" on its own site, and three of them carry a D suffix. Not one of the three walks. G1-D, the newest, is a G1 torso on a telescopic column that travels between roughly 1,260 and 1,680 mm. Unitree counts 19 degrees of freedom outside the end effector — seven per arm, two in the waist, one in the column, two in the base — and offers the base as a choice between a passive castor stand and a powered chassis capped at 1.5 m/s. H2-D does the same thing one size up: no legs, a lifting column with 0.5 m of travel, 135 kg, 7 kg rated payload per arm. R1-D goes furthest and drops the base as well — a torso, a head and two arms bolted to a flat mounting plate, from 4,290 USD in Unitree's own shop. The reason is written into the product rather than announced. G1-D is not sold as a robot at all but as an end-to-end platform, in which the machine is one of three components; the other two are a data-acquisition console built for hundreds of robots collecting simultaneously and a training stack that exports a policy back to the machine in one click, with declared support for the open pi and GR00T families and Unitree's own UnifoLM-WMA-0. For that job legs are a liability: they add cost, they add a balance problem, and a demonstration recorded from a swaying torso is a worse training example than one recorded from a rigid column. The height range does the work that legs would otherwise do, letting the same arms reach a low bench and a high shelf. It is a quiet admission about where humanoid value sits in 2026. The walking machines — G1, H2, R1 — remain the showpieces, and Unitree has spent the summer demonstrating boxing and ping-pong with them in Beijing. The versions the company positions for actual deployment and data work are the ones that gave up the feature the whole category is named for. One caveat on availability: Unitree publishes a price for exactly one of the three, the entry R1-D configuration. G1-D and H2-D are quote-on-request, and none of the three carries a published mass or payload figure on the G1-D page. In the United States, the FCC Covered List entry of 28 July 2026 blocks new equipment authorisations for the manufacturer, which affects new models rather than machines already authorised.
Unitree G1-D →DeepSeek ships the final V4-Pro: open weights, 1M context, agentic scores up sharply
DeepSeek published DeepSeek-V4-Pro-0813 on Hugging Face on 13 August 2026, describing it as the official release of V4-Pro and the successor to the preview weights that had carried the name since April. The repository went up under the MIT licence, so the full model — 1.65 trillion parameters in a sparse mixture-of-experts layout, 61 layers, 384 routed experts with six active per token, shipped in FP8 — can be downloaded and self-hosted. The context window stays at 1,048,576 tokens. The architecture is unchanged from the preview; what is new is a DSpark speculative-decoding module attached to the model and a reworked reasoning control. The `reasoning_effort` parameter now takes three levels — low, high and max — letting callers trade deliberation time against cost. DeepSeek also dropped the Jinja chat template in favour of a documented Python encoder, a change that will require work from anyone running the weights locally. The gains DeepSeek reports are concentrated in agentic and coding work, and they are large. On the company's own table the new build reaches 62.7% on DeepSWE against 12.8% for the preview, 61.5% on NL2Repo against 38.5%, 83.3% on Cybergym against 52.7% and 87.9 on Terminal Bench 2.1 against 72.1. On Humanity's Last Exam it scores 42.7% without tools and 60.0% with them, up from 37.7%. Against rivals in the same table the picture is closer. Kimi K3 edges it on Terminal Bench (88.3) and DeepSWE (67.5), and Anthropic's Fable 5 leads on Humanity's Last Exam (53.3 without tools). What DeepSeek keeps is the position it has held all year: comparable results at frontier level, with the weights published rather than rented. Benchmark figures here are the vendor's own and have not yet been reproduced independently.
DeepSeek-V4-Pro →The US closes its border to new Chinese robots — and BYD's first humanoid walks out three days later
The United States Federal Communications Commission updated its Covered List on 28 July 2026 to include foreign-made humanoid and quadruped robots, alongside connected power inverters. The effect is narrow in wording and wide in practice: the agency will not grant equipment authorisation to new device models, which is the permission a radio-emitting device needs before it can be legally imported or sold in the US. Robots already authorised, and units already bought, are not affected. The stated reasoning came from a White House task force which concluded that foreign-built robots could constitute a cybersecurity risk to critical infrastructure, with espionage and the protection of the domestic AI industry named alongside it. China objected publicly. The measure is not written against a single company, but the market it describes is overwhelmingly Chinese: the great majority of humanoids and almost all commercial quadrupeds shipping today are built in China. Three days later the point was illustrated. On roughly 1 August, at its Di Space experience centre in Zhengzhou, BYD showed the physical unit of Xiao Di, the first humanoid from the world's largest electric-car maker. It stands 1.61 metres and weighs 58.5 kilograms; Chinese reports credit it with 31 degrees of freedom across hands, legs, hips and neck, 360-degree panoramic vision, and real-time interpreting between six Chinese dialects and six foreign languages. Its job is not manufacturing. BYD wants two or three of them in each dealership, welcoming visitors, explaining cars and running product demonstrations, and describes them as support for sales staff rather than a replacement. The company set up the team behind the robot in late 2024, inside its 15th Business Division; it presents Xiao Di as a working machine, not a concept. For a catalogue like this one, the ruling changes what a status badge means. A robot can be in volume production, publicly priced and commercially deployed, and still be unavailable in the largest single market for reasons that have nothing to do with the machine. The near-term winner is the one American humanoid programme with no Chinese competition to price against; the near-term loser is any buyer outside China who was waiting for the cheap option to arrive.
BYD Xiao Di →The model Europe was not allowed to have: Tencent drops the regional carve-outs from Hy3 and goes global
Chinese open-weight models have spent the past two years competing on size and price. Tencent's Hy3 has quietly been competing on something else: who is allowed to download it. When the preview build appeared in April 2026, its licence carried a geographic exclusion — the European Union, the United Kingdom and South Korea were carved out. The official release on 6 July 2026 removed the clause entirely and put the weights under a plain Apache 2.0 licence, one of the most permissive terms yet attached to a model of this scale. On 5 August Tencent completed the move by announcing global availability: the WorkBuddy assistant, the Miora design tool and Tencent Cloud TokenHub, plus third-party access through OpenRouter and downloads on Hugging Face and ModelScope. The model itself is a sparse mixture-of-experts design: 295 billion parameters in total, roughly 21 billion activated per token across 192 experts with top-8 routing, 80 layers, and a separate 3.8-billion-parameter multi-token-prediction stack that speeds up decoding. The context window is 256,000 tokens. The distinctive feature is that the amount of deliberation is a setting rather than a separate product — no_think answers immediately, think_low reasons briefly, think_high runs an extended chain — so one endpoint covers both a chat reply and a long analytical job. Tencent's own framing is that Hy3 performs comparably to flagship models with two to five times as many parameters, across reasoning, instruction following, in-context learning, code generation and agentic work. The company also reports that within a week of launch, API traffic to Hy3 ran 68 times higher than to the previous generation, putting it first on OpenRouter's usage leaderboard. Listed OpenRouter pricing is 0.1288 US dollars per million input tokens and 0.5336 per million output; access through WorkBuddy is free until 31 August 2026. The usage number is the one to watch. Benchmark parity claims from vendors are routine; a 68-fold jump in paid API calls is a statement about how many developers actually switched. Independent coverage broadly agrees with the performance claims, with coding the one area where rival open-weight models still hold an edge.
Tencent Hy3 →Unitree prices its IPO at 150.8 yuan — a humanoid maker valued at 219 times earnings
Unitree Robotics (Yushu Technology) has set the price for its Shanghai STAR Market listing at 150.8 yuan per share, according to the issuance announcement filed with the Shanghai Stock Exchange under code 688836. The company is selling 40,446,434 new shares, about 10 percent of its enlarged capital, which raises roughly 6.1 billion yuan and implies a market capitalisation near 61 billion yuan — some 9 billion dollars. The pricing puts Unitree on a trailing price-to-earnings ratio of 219.23, against a sector average of 38.56 quoted in the same filing. Public subscription opens on 10 August. Entities linked to DeepSeek and Tencent are listed among the strategic investors. Unitree would be the first humanoid-robot maker listed on a mainland Chinese exchange. The company reached that point on the back of quadrupeds such as the Go2 and a humanoid line running from the H1 through the G1 to this year's H2 — products that undercut Western competitors by an order of magnitude on price. The multiple the market has accepted says less about current profits than about how much of the humanoid industry investors expect Unitree to end up owning.
Unitree H2 →Unitree starts pricing its IPO — the first humanoid maker to list on China's A-share market
Unitree Robotics began preliminary price inquiries for its STAR Market flotation on Wednesday 5 August 2026, the step that sets the issue price before subscription opens on 10 August. The offer covers about 40.45 million new shares — 10 percent of the enlarged share capital — with a target of 4.2 billion yuan (roughly 0.6 billion US dollars). China's securities regulator approved the registration on 2 July, so barely a month passed between clearance and pricing. Valuation estimates diverge sharply: investment banks put the company at about 40 billion yuan after listing, while CCB International sees 60 to 100 billion, with an optimistic case of 109 billion. The numbers underneath are what make this more than a financing story. Unitree shipped more than 5,500 humanoid robots in 2025 — first in the world by unit volume — and humanoids brought in 868 million yuan, 51.78 percent of total revenue. A company whose business was quadruped robots is now, by its own accounts, mostly a humanoid company. It will be the first embodied-intelligence firm on the A-share market. That gives the sector something it has lacked: a listed pure-play whose quarterly numbers are public, against which the shipment claims that circulate around this industry can finally be checked. wujec.ai catalogues four Unitree machines: G1, H1 Pro, H2 and the Go2 quadruped.
Unitree G1 →Unitree's three-continent rollout stalls at the US border
Unitree Robotics set out to do something no humanoid maker had done before: launch the same machine commercially on three continents inside three weeks. The H1 Pro went on sale in Europe on 22 July 2026, Asia followed on 5 August with a reported deployment of more than five hundred units into logistics warehouses and manufacturing plants, and North America was scheduled for 12 August. The American leg no longer has a route. On 28 July 2026 the FCC extended its Covered List to foreign-produced advanced robotic devices, which withholds the equipment authorisation a new hardware model needs before it can be imported, marketed or sold in the United States. Reporting indicates no current regulatory pathway for the H1 Pro there. The move follows the Pentagon adding Unitree to its list of Chinese military companies earlier in the summer, which bars Department of Defense contracts. Analysts have also pointed to China's National Intelligence Law, under which a Chinese company can be required to provide data to state intelligence services regardless of where its machines operate. No equivalent European rule blocks the sale of the H1 Pro today. The H1 Pro now has its own profile in the wujec.ai catalogue, alongside the Unitree G1, Go2 and H2.
Unitree H1 Pro →BYD shows Xiao Di, its first humanoid robot, in the showroom
BYD presented its first humanoid robot, Xiao Di, in early August 2026 at the carmaker's Di Space experience centre in Zhengzhou. The machine stands 1.61 m tall and weighs 58.5 kg, and reports from the debut single out how smooth its motion is. Unlike several recent Chinese humanoid reveals, Xiao Di is presented as a working prototype rather than a concept. Its job is retail: greeting visitors, walking them through the electric cars on the floor and running product demonstrations. BYD frames the robot as support for showroom staff, not a replacement, and has signalled that two to three units could eventually be stationed in each of its stores. The move is a short step for BYD, which already builds the batteries, electric motors and power electronics a humanoid needs, and it puts the world's largest EV maker on the same board as Tesla's Optimus programme.
BYD Xiao Di →Alibaba launches Qwen3.8-Max, a 2.4-trillion-parameter flagship
Alibaba released Qwen3.8-Max on 3 August 2026, the largest model the Qwen family has produced so far. It is a sparse mixture-of-experts design: 2.4 trillion parameters in total, of which roughly 95 billion are activated per token, with a context window of up to one million tokens. The model is available worldwide through Alibaba Cloud's Model Studio APIs and through QwenWork, the company's workplace agent platform. Alibaba said full model weights would follow for public download, alongside a smaller Qwen3.8-27B variant aimed at hardware-constrained deployments. On the Arena.AI leaderboard Qwen3.8-Max became the highest-ranked Chinese model for text tasks and placed second globally for vision, putting it in the same bracket as current frontier systems from OpenAI and Anthropic. Alibaba shares rose sharply in Hong Kong on the announcement.
Qwen3.8-Max →Moonshot AI ships Kimi K3, the largest open-weight model ever
Moonshot AI released Kimi K3 on July 16, 2026, open-sourcing the 2.8-trillion-parameter weights under a modified MIT license eleven days later. The mixture-of-experts model activates 104B parameters (16 of 896 experts) per token, adds a 1M-token context window and native image/video understanding, and topped Arena's blind Frontend Code evaluation ahead of Claude Fable 5 — the clearest sign yet that open-weight labs out of China are setting the pace at the very top of the frontier race, not just chasing it. K3 succeeds the original Kimi K2 through the K2 Thinking, K2.5, K2.6 and K2.7-Code line, and developers have already wired it into Cursor and Vercel, running it on consumer GPUs within days of release.
Kimi K3 →DeepSeek's 1.6-trillion-parameter flagship goes GA — and starts charging by the clock
DeepSeek-V4-Pro left preview on 19 July 2026, three months after its first public release. The weights are the same ones published on 24 April: a sparse mixture-of-experts model with 1.6 trillion total parameters, roughly 49 billion of them active per token, a one-million-token context window and an MIT licence that allows anyone to download and self-host it. What general availability changed was the commercial side. DeepSeek introduced peak-time pricing for the first time, doubling rates during Beijing business hours; off-peak the API costs 0.435 US dollars per million input tokens and 0.87 per million output. Cached input is charged at roughly a hundredth of the off-peak input rate. On 24 July the company retired the legacy deepseek-chat and deepseek-reasoner endpoints, so integrations now have to name a model explicitly. On DeepSeek's own GA figures the model scores 80.6% on SWE-bench Verified — the strongest published result for an open-weights model — with a Codeforces rating of 3,206, 57.9% on SimpleQA Verified and 37.7% on Humanity's Last Exam. Independent measurement tells a subtler story: Artificial Analysis rated V4-Pro at 44 on its Intelligence Index at GA, down from 52 at preview. The model did not get worse; the field moved during the three months it spent in preview. The smaller DeepSeek-V4-Flash and the earlier DeepSeek-R1 remain available, and all three now have profiles in the wujec.ai catalogue.
DeepSeek-V4-Pro →