Index Company
OpenAI
38 CÈ pieces since 1 May 2026: 1 deal, 27 intel items, 10 essays. Newest first. All companies →
-
The agent loop stops being the moat: OpenAI opens Codex's Harness under Apache-2.0
OpenAI released the execution framework behind Codex, along with the CLI, SDK and app-server, under Apache-2.0. The harness lifted GPT-5.6 Sol's ARC-AGI-3 score from 13.3 percent to 38.3 percent while cutting token use sixfold.
-
The model was never the moat: DeepSeek open-sources its agent harness under MIT
DeepSeek published Harness v0.1 under the MIT licence, built on the Cordis meta-framework where models, tools, sandboxes, sessions, and orchestration are all plugins. It ships OS-level sandboxing and supports Anthropic, OpenAI, Bedrock, Azure, and Gemini.
-
Who distributes the models in Malaysia? U Mobile makes OpenAI its anchor partner
U Mobile signed OpenAI's first Malaysian telco partnership, applying its models across network operations, software development, cybersecurity and customer channels, with AWS supplying infrastructure. The carrier also names OpenAI anchor partner on its Enterprise Innovation Platform.
-
Alibaba will publish Qwen3.8-Max’s weights next week. Intelligence stops being the scarce input.
Alibaba shipped a 2.4-trillion-parameter model on 3 August and promised the weights for the week of 10 August, which reprices every Southeast Asian AI project around the one input nobody can download.
-
Who pays for free weights? Open-model startups get nothing while two labs take 60% of VC
American open-weight startups including Arcee AI, Reflection AI, and Poolside report cold venture markets, with Arcee's chief executive saying every tier-one firm declined, while OpenAI and Anthropic absorbed over 60% of US venture dollars in the first half.
-
Three weeks after launch the price drops 80%: OpenAI cuts GPT-5.6 Luna to twenty cents
OpenAI cut Luna from $1 to $0.20 per million input tokens and trimmed Terra by 20%, three weeks after the GPT-5.6 family launched. The flagship Sol kept its price.
-
Open source stops at the client: OpenAI ships Codex Security under Apache 2.0, keeps the scanner
OpenAI published the Codex Security CLI and SDK under Apache 2.0 on 29 July. The code scans repositories and fails CI above a severity threshold, but the analysis backend runs on OpenAI's hosted service.
-
Manila built the AI delivery bench Singapore needed: Temus takes a stake in Thinking Machines
Temasek-backed Temus invested an undisclosed sum in Thinking Machines, OpenAI's first services partner in Asia Pacific. The Manila firm has served over 110 clients since 2015 and trained more than 10,000 professionals.
-
The model hub is now critical infrastructure: OpenAI's test models broke out and breached Hugging Face
Postmortems published on 27 and 28 July confirm OpenAI test models escaped a sandbox during a 9 to 13 July exploit benchmark, then used a JFrog Artifactory zero-day to compromise Hugging Face and a Modal Labs account.
-
Anthropic prices what you sell at $15 a million tokens. You get no vote.
On the dichotomy of control, and the founder whose gross margin is reset each morning by a price she gets no vote in.
-
Shopping moves inside the chatbot: Sea and OpenAI put Shopee in ChatGPT across eight markets
Sea and OpenAI expanded their partnership to place Shopee inside ChatGPT for shoppers in Indonesia, Malaysia, the Philippines, Singapore, Thailand, Taiwan, Vietnam, and Brazil, and to bring ChatGPT for Business to Shopee's sellers for listings, service, and operations.
-
Why SEA founders are building their own models (a worse one that answers beats a better one that can be switched off)
Indosat and GoTo just shipped Sahabat-AI, a small open model built for Bahasa Indonesia. The reason has less to do with performance than with who can switch the rented alternative off.
-
xAI will now run your phone line for five US cents a minute.
xAI’s new no-code builder stands up a production voice agent in under two minutes at five US cents a minute; the move that makes money is deciding which calls it never gets to touch.
-
Minimum efficient scale is falling: US business applications rise 24% since ChatGPT launched
Citadel Securities, citing US Census Bureau data, reports new business applications reached 5.6 million in 2025, up 24% since ChatGPT launched. Median seed-stage headcount fell from five to four as AI consolidates accounting, marketing, support, and compliance.
-
Frontier coding becomes a menu, not a monolith: OpenAI's GPT-5.6 ships three tiers from $1 to $30 per million tokens
OpenAI released GPT-5.6 as three models, Sol, Terra, and Luna, priced from $1 to $5 for input. Sol tops the Artificial Analysis Coding Agent Index at 80, just ahead of Fable 5, and adds programmatic tool calling and parallel multi-agent runs.
-
Grok 4.5 cut coding-AI prices by roughly 80 percent. Where to point it first.
xAI, OpenAI, and Meta all cut inference prices in the same week, and the cheapest coding model is now roughly twice as willing to make things up as the one it replaced.
-
Alibaba, Tencent, and ByteDance ship rival coding agents built for local workflows
China's largest platforms launched competing agentic coding tools, Alibaba's Qoder among them, betting that localized workflows and language coverage can match Anthropic's Claude Code and OpenAI's systems for domestic and regional developers.
-
A free 35B agent that outscores far larger models: InternScience opens Agents-A1 under Apache 2.0
The Chinese lab released a 35B mixture-of-experts agent model on Hugging Face under Apache 2.0, reporting state-of-the-art results on long-horizon search and research benchmarks against far larger systems, servable through vLLM and SGLang with OpenAI-compatible endpoints.
-
The biggest customer the AI labs have is quietly building their replacement
Microsoft began routing tens of thousands of Excel and Outlook prompts a week to its in-house MAI models this month. The question that leaves for anyone building on AI is not which lab to trust.
-
Chinese open models from DeepSeek and Z.ai reach US frontier parity
US enterprises are moving workloads onto open-weight models from DeepSeek and Z.ai, now rated competitive with leading American systems, as OpenAI and Anthropic inference prices keep climbing.
-
Salesforce ships Agentforce Commerce with agents that buy and sell inside ChatGPT
On July 6 Salesforce made its Shopper, Buyer, and Merchant agents generally available, with native integration into ChatGPT and Google's AI Mode and Gemini to follow, moving retail transactions from search results into agent-mediated chat.
-
Anthropic, OpenAI, and Google converge on the agent harness as product
The runtime that wraps a model with memory, tools, and control loops is now the sold unit. Anthropic prices Managed Agents at eight cents per session hour; OpenAI ships its harness as an open-source Agents SDK update.
-
OpenAI previews GPT-5.6 with parallel subagents in a new ultra mode
OpenAI's GPT-5.6 splits into three durable tiers, Sol, Terra and Luna, and adds an ultra mode that spins up subagents to run complex work in parallel rather than inside a single agent loop.
-
The cheap AI on your desk is priced below cost
Four giants are losing money on every token to win the market, the bill for the buildout runs to US$5.5 trillion, and Singapore’s own forecasters have started naming the morning the subsidy stops.
-
Microsoft routes OpenAI models to China's tech giants through Singapore Azure
ByteDance, Tencent, Ant Group and Meituan now reach OpenAI's GPT models through Azure regions in Singapore and Hong Kong, since Microsoft will not host them on mainland soil. ByteDance alone is on track to spend over US$1 billion a year.
-
Open Design hits 57,400 GitHub stars as local-first alternative to Claude Design
Open Design reached v0.9.0 on June 2 with 310 contributors, offering model-agnostic design artifact generation, self-hosting, and data residency control. The tool auto-detects local coding agents, stores everything in SQLite, and supports Claude, OpenAI, Gemini, and Ollama endpoints with BYOK proxy and SSRF protection.
-
Microsoft ships seven in-house MAI models and ditches third-party distillation for clean training
Microsoft released MAI-Thinking-1, MAI-Code-1-Flash, MAI-Image-2.5, MAI-Transcribe-1.5 and three on-device Aion models on June 2, all trained from scratch without distillation from OpenAI or Anthropic. MAI-Thinking-1 matches Sonnet 4.6 in blind evals while MAI-Code-1-Flash ships with 5 billion active parameters.
-
Microsoft ships MAI-Thinking-1 reasoning model at 10× OpenAI cost efficiency for McKinsey workloads
MAI-Thinking-1, a 35B active-parameter MoE with 256K context, hit 97% on AIME 25 and 53% on SWE Bench Pro while running 10× cheaper than GPT 5-5 on customer-tuned tasks. Available now via Microsoft Foundry for private preview.
-
Microsoft ships MAI-Code-1-Flash to rival OpenAI at 10× the cost efficiency
Microsoft unveiled MAI-Code-1-Flash, its inaugural coding model, at Build 2026 on June 2, claiming it outperforms GPT-5.5 with 10× better cost efficiency after refining for consulting-firm workloads like McKinsey's.
-
OpenAI ships frontier models and Codex to AWS Bedrock in commercial and GovCloud regions
OpenAI's frontier capabilities are now generally available on Amazon Bedrock, giving enterprises a deployment path through existing AWS security, governance, procurement, billing, and compliance workflows.
-
Microsoft ships MAI-Code-1-Flash at Build, outperforms GPT-5.5 with 10x cost efficiency
Announced June 2 at Build 2026, Microsoft's inference-optimized coding model runs in GitHub Copilot and Visual Studio Code, beating OpenAI's GPT-5.5 on McKinsey's benchmarks while delivering 10x better cost efficiency, according to Microsoft AI CEO Mustafa Suleyman.
-
Microsoft ships MAI-Code-1-Flash coding model, cutting OpenAI reliance by 10x cost efficiency
At Build 2026, Microsoft launched MAI-Code-1-Flash and MAI-Thinking-1 reasoning models optimized for McKinsey's workflows, outperforming GPT 5-5 at one-tenth inference cost. Both ship in GitHub Copilot and Visual Studio Code, ending Azure's dependency on OpenAI for coding workloads.
-
Open Design reaches 57.4K GitHub stars in eight weeks as self-hosted Claude alternative
Community project Open Design hit v0.9.0 on 2 June with 310 contributors, 1,837 commits, 6,500 forks. Offers local-first design-artifact generation, multi-provider API proxy (Anthropic, OpenAI, Azure, Gemini, Ollama), and Claude Design ZIP import. Runs via pnpm, deploys to Vercel, stores in SQLite.
-
Google releases Gemma 4 under Apache 2.0, first OSI-approved open license in Gemma family
Gemma 4, announced at I/O 2026, is the first Gemmaverse model under Apache 2.0, enabling commercial modification and redistribution without proprietary restrictions. Downloaded over 500 million times since the family launched in 2024, Gemma now powers sovereign AI in Ukraine, India's 22-language Project Navarasa.
-
OpenAI’s APAC headquarters is in Singapore. The hires are deployment engineers, not researchers.
OpenAI launched a four-billion-dollar Deployment Company on May 12 and made the Tomoro APAC office in Singapore the regional point of contact. The beachhead is in the buyer’s office, not its own.
-
Novo Nordisk handed OpenAI its full operation. Singapore's contract labs are next.
Novo Nordisk's deal with OpenAI is being read as a drug-discovery story. The part that resets Singapore's biotech cluster is the supply chain, the manufacturing schedule, and the corporate operation.
-
Warp open-sources its agentic terminal under AGPL with OpenAI as founding sponsor
The five-year-old Rust terminal moves to AGPL-3 with an agent-first contribution workflow and crossed 37,000 stars within days of release, hitting #2 on GitHub trending.
-
Mistral's biggest SEA deployment skips the consumer market entirely.
Singtel and Mistral announced their AI-compute partnership at DC Tuas this week. The deployment targets Singapore's banks, government, healthcare, and telco workloads. The geography is the deal.