OpenAI Product Pulse

DeepSeek-V4 Flash achieves 80% of Luna performance at 1/6 cost. Anthropic Opus 5 beats Sol on ARC-AGI-3. Microsoft replacing OpenAI models in Copilot/Excel/Outlook with own MAI models (85% cost savings). OpenAI slashed Luna prices 80%, Terra 20%, introduced Sol Fast mode. Cost pressure intensifies as OpenAI remains unprofitable despite 1B weekly users. GPT-5.6 Sol hit #1 on ProgramBench. SpaceXAI Grok 4.6 matches Sol on some benchmarks, undercuts price. Ramp data shows Sol generating more spend than Fable 5, but Anthropic leads overall business adoption 43.5% vs 39.7%. ChatGPT's web share dropped from 76% to 54% as Gemini and Claude gain ground. Benchmark saturation article highlights convergence of top models. Comprehensive model comparison reveals 13x pricing gap between GPT-5.6 tiers vs Claude Fable 5 and Gemini 3.6/3.7 Flash. Anthropic has overtaken OpenAI on revenue, with $65B run rate vs OpenAI's $40B, driven by agent-first strategy (Claude Code, Cowork). This confirms enterprise agents as the revenue engine and raises questions about OpenAI's consumer bets. OpenAI and Anthropic restrict access to strong models, potentially slowing enterprise adoption and shifting competitive dynamics. OpenAI's revenue growth slowing to 18% QoQ while Anthropic surges, raising profitability concerns. New: OpenAI offers 50% off GPT-5.6 lineup on OpenRouter to challenge Anthropic, driving developer migration from Claude. New Ramp data shows OpenAI clawing back business share after losing to Anthropic in May, driven by GPT-5.6 Sol and Anthropic's data retention requirement. Q2 financials confirm Anthropic leading revenue growth. New: Claude Opus 4.8 beats GPT-5.3 Codex on overall benchmarks (76.2 vs 65.66) but GPT-5.3 Codex is significantly cheaper, reinforcing cost war dynamics. New: Upcoming model releases (Astra/GPT-6, Fable 5.1, Grok 4.7, Kimi 3.5) signal intensifying competition. New: Ramp data confirms OpenAI regaining enterprise market share over Anthropic, driven by GPT-5.6 Sol and Anthropic's data retention policy backfiring. New: AT&T routing 40% of AI usage to open models with only 2% quality drop, directly challenging OpenAI's enterprise pitch and highlighting the cost advantage of open-source alternatives. New: OpenAI cuts developer pricing for GPT-5.6 Sol by more than 20% on API, further intensifying cost wars. New: Microsoft gave up OpenAI exclusivity and built its own reasoning model, now a direct competitor. This is a major shift in the partnership dynamic, as Microsoft was OpenAI's largest investor and cloud partner. The new model competes directly with GPT-5.6 and could erode OpenAI's enterprise foothold. IBM partnership may help OpenAI counter competitive pressure by deepening enterprise relationships. New: A mysterious model called OX Alpha (likely Zhipu AI's GLM-5.x) is reportedly beating GPT-5.6 Sol and Claude Fable 5 at coding (80% DeepSWE), with 1M context and free preview on OpenRouter. This adds a new competitive threat to OpenAI's coding dominance and could accelerate pricing wars. New: Inherent's Faraday agent outperforms OpenAI and Anthropic on research replication using a much smaller model, adding a new competitor threat in specialized agents. New: Legal AI vendors like Harvey and Thomson Reuters are building their own models to cut inference costs, threatening OpenAI's enterprise revenue. OpenAI is countering by hiring legal execs and deploying ChatGPT Enterprise at Willkie Farr. New: Local AI tools (LM Studio, Ollama, Jan, GPT4All, AnythingLLM) are replacing ChatGPT for some users, challenging subscription stickiness and validating the cost war narrative. New: OpenAI cuts GPT-5.6 Sol API prices by up to 33% through November 21, a temporary but aggressive move to maintain market share. Article confirms enterprise cost sensitivity is reshaping AI adoption, with Anthropic's flagship lagging while cheaper models win. Ramp data showing Anthropic overtaking OpenAI in spending share is notable. New: Open models token share rising dramatically (Vercel chart: 28%→62% in two months), pressuring OpenAI's pricing and enterprise strategy. New: Ramp data shows enterprise AI loyalty fading — Anthropic leads paid adoption 43.5% vs 39.7%, 52% use both, confirming multi-vendor trend. New: Hands-on comparison of GPT-5.6 vs Claude Opus 5 vs Grok 4.6 on a complex web project shows GPT-5.6 solid but with UX flaws. **Latest:** Altman hints at secret math model for enterprise, potentially a competitive response to cost pressure.