https://www.userightai.com Pricing changes, new models, and recommendation updates from UseRightAI. Independent. No paid rankings. Updated weekly. en-us Sun, 11 Oct 2026 00:00:00 GMT hello@userightai.com (UseRightAI) hello@userightai.com (UseRightAI) 1440 https://www.userightai.com/models/gemini-3-1-pro chg_correction_context_windows_2026_10_11 Two long-standing errors in our catalog are fixed. Gemini 3.1 Pro was listed with a 2M-token context window; Google's Vertex AI documentation and the Vercel AI Gateway both list 1M. Grok 4 was listed at $2/$6 with a 2M window; it launched in July 2025 at $3/$15 per million tokens with 256K, and xAI retired it from its API on May 15, 2026. Several pages recommended Gemini 3.1 Pro for long documents because of the 2M figure. With every current frontier model at about a million tokens, those recommendations now rest on reasoning quality and price instead — the long-documents guide points to GPT-6.1 Sol, and research and PDF guides to Claude Opus 5.5 and Gemini 3.8 Flash. Comparison pages that declared Gemini or Grok the context-window winner have been corrected, and the 2026 timeline no longer lists Claude Haiku 4.5, Llama 4 or Grok 4, which were 2025 releases. Corrected October 11, 2026. Sun, 11 Oct 2026 00:00:00 GMT recommendation https://www.userightai.com/models/gpt-5-6-sol chg_correction_gpt_5_6_sol_price_2026_10_10 Our GPT-5.6 Sol pages showed $2 per million input tokens and $10 per million output. That was wrong. GPT-5.6 Sol launched at $5/$30, and since late August 2026 OpenAI has charged a promotional $4/$20, which OpenAI says runs through at least November 21, 2026. We corrected the price on every page that quoted it, along with comparisons that called GPT-6 Astra 'five times' Sol's price — against the real rate it is two and a half times. The $2/$10 figure belongs to GPT-6 Sol and GPT-6.1 Sol. Corrected October 10, 2026 against the Vercel AI Gateway catalog and OpenAI's model page. Sat, 10 Oct 2026 00:00:00 GMT pricing https://www.userightai.com/models/claude-haiku-5-5 chg_claude_haiku_5_5_2026_10_07 Anthropic made Claude Haiku 5.5 generally available on October 7, 2026, after announcing it with Sonnet 5.5 on September 28. It is the first Haiku with adjustable reasoning effort, from off to max. Pricing is tiered: $0.10 per million input tokens and $0.50 per million output up to 100K input tokens, and $0.50/$2.50 above that. It has a 1M-token context and 128K output. Anthropic reports 39.2% on Terminal-Bench 4.0 at max effort (12.8% at low effort) and 72.4% on the offline subset of OSWorld 2.1, against 15.7% for Haiku 4.5. Artificial Analysis scores it 43 on its Intelligence Index, ahead of Gemini 3.8 Flash (41) and GPT-6 Luna (38). Verified October 10, 2026. Wed, 07 Oct 2026 00:00:00 GMT new-model https://www.userightai.com/models/mistral-large-4 chg_mistral_large_4_2026_10_06 Mistral opened a hosted preview of Mistral Large 4 on October 6, 2026: a natively multimodal mixture-of-experts model with about 1 trillion total parameters, priced at $0.68/$2.09 per million tokens, with a 512K context and up to 256K output. Mistral says the open weights follow at the end of October. Every benchmark figure so far is Mistral's own; Vals AI's independent index places it 32nd of 44 models (48.05%). We have added it to the directory with that caveat. Verified October 10, 2026. Tue, 06 Oct 2026 00:00:00 GMT new-model https://www.userightai.com/models/gpt-6-1-sol chg_gpt_6_1_sol_2026_09_29 OpenAI released GPT-6.1 Sol on September 29, 2026 at DevDay, one week after GPT-6 Sol. OpenAI says it comes close to GPT-6 Astra on agentic coding, computer use and professional work at one fifth of Astra's price, and Artificial Analysis puts it one point below Astra on its Intelligence Index. Pricing is $2/$10 per million tokens up to 272K input tokens ($4/$15 above), with cached input at $0.10; a Flex tier costs $1/$5 and an Ultrafast tier $12/$60. It has a 1.05M-token context and 128K output. At launch it reached Plus, Pro, Business, Enterprise and Edu users through ChatGPT Work and Codex rather than ordinary ChatGPT chats. OpenAI did not release a GPT-6.1 Astra. Verified October 10, 2026. Tue, 29 Sep 2026 00:00:00 GMT new-model https://www.userightai.com/models/claude-sonnet-5-5 chg_claude_sonnet_5_5_2026_09_28 Anthropic released Claude Sonnet 5.5 on September 28, 2026, positioned for well-scoped everyday work such as building features, fixing bugs and producing documents, slides and spreadsheets. It costs $2/$10 per million tokens — less than Sonnet 5's $3/$15 — with cache reads halved to $0.10, a 1M-token context and 128K output. In Anthropic's launch table it scores 70.6% on Terminal-Bench 4.0 (Opus 5.5: 66.4%), 80.1% on OSWorld 2.1 (81.8%) and 64.5% on Humanity's Last Exam (67.7%). Anthropic says it generates more than 30% faster than Sonnet 5 and still recommends Opus 5.5 for long, loosely defined tasks. Verified October 10, 2026. Mon, 28 Sep 2026 00:00:00 GMT new-model https://www.userightai.com/models/claude-opus-5-5 chg_claude_opus_5_5_2026_09_22 Anthropic released Claude Opus 5.5 on September 22, 2026. Its launch table reports 89.9% on SWE-bench Pro, ahead of Claude Fable 5.1 (81.2%) and Opus 5 (79.2%), and 93.9% on SWE-bench Multilingual. The price fell from Opus 5's $5/$25 to $4/$20 per million tokens, with cache reads at $0.20, and Anthropic says it spends fewer tokens per finished task. It is not ahead on every row: Opus 5 still scores higher on Toolathlon Verified (80.6% vs 77.8%), and its Terminal-Bench 4.0 result depends on the effort setting (66.4% at xhigh, 64.8% at max). Context is 1M tokens with 128K output; Fast mode costs $8/$40. Opus 5.5 is now the top coding pick across the directory. Verified October 10, 2026. Tue, 22 Sep 2026 00:00:00 GMT new-model https://www.userightai.com/models/gpt-6-sol chg_gpt_6_sol_luna_2026_09_22 OpenAI released GPT-6 Sol and GPT-6 Luna on September 22, 2026, about three weeks after GPT-6 Astra. Sol costs $2/$10 per million tokens and Luna $0.10/$0.50, both with a 1.05M-token context and 128K output. The release is mainly a price cut: OpenAI's own charts show Sol at 33.2% on AutomationBench for about $0.27 per task, but below GPT-5.6 Sol on DeepSWE v1.1 (68.8% vs 72.7%) and OSWorld 2.0 (64.4% vs 66.2%). GPT-6.1 Sol replaced Sol a week later at the same price. Verified October 10, 2026. Tue, 22 Sep 2026 00:00:00 GMT new-model https://www.userightai.com/models/grok-4-7 chg_grok_4_7_2026_09_21 Grok 4.7 was released on September 21, 2026 at Grok 4.6's $2/$6 per million tokens, with a 500K-token context. The vendor's table reports 46.3% on CursorBench 4.0 and 37.6% on Terminal-Bench 4.0, still behind Claude Fable 5.1 on the coding rows. Snorkel AI's Senior SWE-Bench puts it at 40.0% pass@3 (Grok 4.6: 38.9%). Independent SWE-Together runs measured roughly twice Grok 4.6's tokens per task, and its maintainers caught the model working around their sandbox. No SWE-bench Verified or Pro figure was published. Verified October 10, 2026. Mon, 21 Sep 2026 00:00:00 GMT new-model https://www.userightai.com/models/deepseek-v4-1-flash chg_deepseek_v4_1_flash_2026_09_10 DeepSeek released V4.1 Flash on September 10, 2026, took the previous V4 Flash endpoints offline the same day, and from September 14 began serving all V4-Pro API requests with V4.1 Flash at Flash rates until a V4.1-Pro ships. The model is natively multimodal, has a 1M-token context and adjustable reasoning effort, and its weights are on Hugging Face under the MIT licence. DeepSeek says it beats V4-Pro on coding and agent tasks; those figures are self-reported. Hosted prices vary by provider; the Vercel AI Gateway lists $0.30/$1.20 per million tokens. Verified October 10, 2026. Thu, 10 Sep 2026 00:00:00 GMT new-model https://www.userightai.com/models/gpt-6-astra chg_gpt_6_astra_2026_09_03 OpenAI released GPT-6 Astra on September 3, 2026, two days after Anthropic's Claude Fable 5.1, calling it its most intelligent and aligned model. The headline is computer use: 72.6% on OSWorld 2.0 in roughly 47% less time per task than GPT-5.6 Sol (65.7%), plus 92.7% on ScreenSpot-Pro. On agentic coding it posts 57.9% on Terminal-Bench 4.0, ahead of Fable 5.1 (55.8%), Opus 5 (52.3%) and Sol (37.3%), and on frontier math it saturates FrontierMath Tier 4 at 97.6% against Fable 5.1's 87.8%. It is not a clean sweep: Fable 5.1 still leads Humanity's Last Exam with tools (65.0% vs 57.2%) and the Artificial Analysis Intelligence Index (65.7 vs 61.2), and OpenAI published no SWE-bench figure. Standard API pricing is $10 per million input tokens and $50 per million output — five times GPT-5.6 Sol and identical to Fable 5.1 — with a Fast mode at 2x speed for 2x price. Model docs list a 1,050,000-token context, 128,000 max output and an April 30, 2026 knowledge cutoff. Availability: rolling out over the coming days to ChatGPT Plus, Pro, Business and Enterprise (inside existing allowances, Enterprise off by default, a GPT-6 Astra Pro tier for Pro/Business/Enterprise), the OpenAI API as gpt-6-astra, Microsoft Azure and Amazon Bedrock. Astra meets the Critical cybersecurity threshold under OpenAI's Preparedness Framework — 100% on ExploitBench — so advanced cyber tasks are refused outside the OpenAI Daybreak program, and OpenAI notes its written reasoning is harder to monitor than Sol's. Every figure is from OpenAI's launch post and model documentation, verified September 4, 2026. Thu, 03 Sep 2026 00:00:00 GMT new-model https://www.userightai.com/models/gemini-3-8-flash chg_gemini_3_8_flash_2026_09_02 Google released Gemini 3.8 Flash on September 2, 2026, its third Flash model in about six weeks, at the same $0.75/$3.75 per million tokens as 3.7 Flash. Google reports 90.8% on Terminal-Bench 2.1 (3.7 Flash: 81.6%) and 54.9% on HLE-Verified. Artificial Analysis measured about 302 output tokens per second but also found it verbose, using 120M output tokens to run its index against a 71M median — so cost per task runs above the per-token price. It accepts text, images, PDF and video, with a 1M-token context and 65K output. A Gemini 3.8 Flash Cyber variant is limited to governments and vetted partners. Verified October 10, 2026. Wed, 02 Sep 2026 00:00:00 GMT new-model https://www.userightai.com/models/muse-spark-1-3 chg_muse_spark_1_3_2026_09_02 Meta released Muse Spark 1.3 on September 2, 2026 at the original Muse Spark's $1.25/$4.25 per million tokens. Meta describes higher first-attempt accuracy, more reliable tool calling and native understanding of video, images and documents across a 1M-token context. Meta published no standard benchmark table for this point release, so we have not changed its ranking against other providers' models. Verified October 10, 2026. Wed, 02 Sep 2026 00:00:00 GMT new-model https://www.userightai.com/models/claude-fable-5-1 chg_claude_fable_5_1_2026_09_01 Anthropic shipped Claude Fable 5.1 and Claude Mythos 5.1 on September 1, 2026 — the first update to the Mythos-class line since Fable 5 launched on June 9. Base pricing did not move ($10 in / $50 out per 1M tokens), but cache reads were cut 75%, from $1.00 to $0.25 per 1M. Anthropic puts the net effect at roughly 25% cheaper on typical workloads and up to 45% on heavily agentic ones — an upgrade that costs less than the model it replaces. The headline result is Terminal-Bench-Science 0.1, where Fable 5.1 scores 52.6% against Fable 5's 24.7%, Opus 5's 29.0% and GPT-5.6 Sol's 22.4%. It also leads Terminal-Bench 4.0 agentic coding (55.8% vs 52.3% for Opus 5), CursorBench 3.2.0 (73.4%) and AutomationBench (31.4%, nearly double Fable 5's 17.1%). Anthropic published no SWE-bench Verified or Pro figure at launch, so our SWE-bench leaderboard is unchanged; independent evaluator Vals AI ranks Fable 5.1 first of 51 models on its Vals Index, first on LiveCodeBench (90.5%) and first on MMLU Pro (92.4%). Mythos 5.1 is the same model with safeguards lifted, scores 60.9% on Terminal-Bench 4.0, and stays restricted to vetted US cybersecurity and life-sciences organisations. API ID claude-fable-5-1, generally available on the Claude API, Amazon Bedrock, Google Cloud and Microsoft Foundry. Every figure verified against Anthropic's launch numbers and two independent write-ups on September 1, 2026. Tue, 01 Sep 2026 00:00:00 GMT new-model https://www.userightai.com/changes chg_late_august_wave_2026_08_27 August 2026 was the fastest release month on record, and the second half of it produced six models worth ranking. Gemini 3.7 Flash ($0.75/$3.75) lands 80.8% on SWE-bench Verified — Google's best coding score per dollar, though that is introductory pricing that doubles on January 1, 2027. Grok 4.6 ($2/$6) finishes agent tasks in roughly half the turns of its rivals. Z.ai shipped twice: GLM-5.3 ($1.40/$4.40) takes a narrow CyberGym lead over Claude Mythos 5, and GLM-5.3 Flash ($0.15/$0.50) is a 320B-A18B natively multimodal model with MIT weights and a 1M context. Alibaba's Qwen 3.8 Flash ($0.16/$0.47) posts SWE-bench Pro 62.5 from 6B active parameters. And Meta returned to a permissive licence with Muse Glimmer 30B — Apache 2.0, 76.0% SWE-bench Verified, runs on a 24GB GPU. Llama 5 stays out of the ranked catalog: still no published weights, licence or hosted price. Every figure verified against the Vercel AI Gateway live price list and two independent benchmark leaderboards on August 27, 2026. Thu, 27 Aug 2026 00:00:00 GMT new-model https://www.userightai.com/changes chg_2026_midyear_wave_2026_08_06 The directory now covers the full mid-2026 release wave: Claude Opus 5 ($5/$25, new SWE-bench Verified leader at ~96%) and Claude Sonnet 5 ($3/$15, promo $2/$10 through Aug 31); the GPT-5.6 family — Sol ($5/$30), Terra ($2/$12), and Luna ($0.20/$1.20 after its ~80% July 30 price cut); Gemini 3.5 Flash, 3.6 Flash, and 3.5 Flash-Lite; DeepSeek V4-Pro and V4-Flash (open weights, from $0.14/1M); Grok 4.5; Mistral Medium 3.5; Meta's Muse Spark; and — new to the directory — the leading Chinese labs: Alibaba (Qwen 3.7 Max, Qwen 3.8 Max), Moonshot (Kimi K3, Kimi K2.7 Code), and Z.ai (GLM-5.2). Every price and benchmark verified against official sources on August 6, 2026. Thu, 06 Aug 2026 00:00:00 GMT new-model https://www.userightai.com/models/gpt-5-6-sol chg_gpt_5_6_ga_price_cuts_2026_07_30 The GPT-5.6 family (Sol, Terra, Luna) reached full public release on July 9, 2026 after becoming the first frontier models to clear a customer-by-customer US government review. On July 30, OpenAI cut prices: Terra from $2.50/$15 to $2/$12 (~20% off), and Luna from $1/$6 to $0.20/$1.20 (~80% off). Sol stays at $5/$30. All three now have full model pages in our ranked catalog. Thu, 30 Jul 2026 00:00:00 GMT pricing https://www.userightai.com/changes chg_gemini_3_5_flash_cyber_2026_07_21 Google's cybersecurity-specialized fine-tune of Gemini 3.5 Flash launched July 21, 2026, restricted to governments and trusted partners via the CodeMender pilot. It found 55 unique confirmed issues in the V8 engine vs 47 for base 3.5 Flash. Because it has no public access or published API pricing, it stays out of our ranked catalog — same policy we applied to GPT-5.6 during its preview. Tue, 21 Jul 2026 00:00:00 GMT new-model https://www.userightai.com/models/gpt-5-6 chg_gpt_5_6_pricing_published_2026_07_01 OpenAI published official GPT-5.6 API pricing on July 1, 2026. Per 1M tokens: Sol costs $5 input / $30 output (matching GPT-5.5), Terra costs $2.50 / $15 (exactly half of GPT-5.5), and Luna costs $1 / $6. Terminal-Bench 2.1 scores are also out: Sol 88.8% (91.9% in Ultra mode, a new state of the art), Luna 84.3%, Terra 82.5% — vs 88.0% for both GPT-5.5 and Claude Mythos 5. Access is still a limited preview (~20 partner orgs); context windows and the wider benchmark suite remain unpublished, so GPT-5.6 stays out of our ranked catalog until general availability. Wed, 01 Jul 2026 00:00:00 GMT pricing https://www.userightai.com/models/gpt-5-6 chg_gpt_5_6_announced_2026_06_26 OpenAI announced GPT-5.6 on June 26, 2026: three models — Sol (its strongest model yet, with a new "max" reasoning effort and an "ultra" sub-agent mode), Terra (matches GPT-5.5 quality at roughly half the cost), and Luna (its fastest, lowest-cost model). Sol sets a new state of the art on Terminal-Bench 2.1, with agentic gains in coding, biology, and cybersecurity. It launched as a limited preview to select partners, with ChatGPT, Codex, and API access rolling out in the coming weeks. Official pricing, context windows, and benchmark scores have not been published yet — we will post verified numbers as OpenAI confirms them. Fri, 26 Jun 2026 00:00:00 GMT new-model https://www.userightai.com/models/claude-fable-5 chg_claude_fable_5_mythos_5_added_2026_06_09 Anthropic released Claude Fable 5 and Claude Mythos 5 on June 9, 2026, two months after Mythos rocked Wall Street in private preview. Fable 5 — the generally available, Mythos-class model — scores 80.3% on SWE-Bench Pro, an 11-point leap over Opus 4.8's 69.2% and the largest single-release jump of 2026. It posts 1932 on GDPval-AA, has a 1M-token context, and is free for everyone until June 22. Mythos 5 is the same model with safeguards lifted, restricted to vetted partners. Pricing is $10/$50 per 1M tokens. Tue, 09 Jun 2026 00:00:00 GMT new-model https://www.userightai.com/models/claude-fable-5 chg_claude_fable_5_coding_rec_2026_06_09 Claude Fable 5 now leads all premium coding and research recommendations, replacing Opus 4.8 at the top. At 80.3% SWE-Bench Pro it is the clear benchmark leader. Opus 4.8 remains the best-value premium pick at half the price, and Sonnet 4.6 stays the everyday default. Tue, 09 Jun 2026 00:00:00 GMT recommendation https://www.userightai.com/models/claude-opus-4-8 chg_claude_opus_4_8_added_2026_05_27 Anthropic launched Claude Opus 4.8 on May 27, 2026. It scores 69.2% on SWE-Bench Pro (up from 64.3% on Opus 4.7), 88.6% on SWE-Bench Verified, and 1890 Arena Elo — 121 points ahead of GPT-5.5. Pricing is unchanged at $5/$25 per 1M tokens. Key new feature: native parallel subagents. Wed, 27 May 2026 00:00:00 GMT new-model https://www.userightai.com/models/claude-opus-4-8 chg_claude_opus_4_8_coding_rec_2026_05_27 Claude Opus 4.8 now leads all premium coding recommendations, replacing Opus 4.7. At 69.2% SWE-Bench Pro vs 64.3%, it is the clear benchmark leader. Opus 4.7 is retained for legacy comparisons and pinned integrations. Wed, 27 May 2026 00:00:00 GMT recommendation https://www.userightai.com/models/gpt-5-5 chg_gpt_5_5_added_2026_04_26 OpenAI's GPT-5.5 is now included in model rankings with SWE-Bench Pro, Terminal-Bench, pricing, and context-window data. Sun, 26 Apr 2026 00:00:00 GMT new-model https://www.userightai.com/models/claude-opus-4-7 chg_opus_4_7_ranking_2026_04_26 Claude Opus 4.7 now leads premium coding recommendations because it has the strongest current public SWE-Bench Pro score among generally available models. Sun, 26 Apr 2026 00:00:00 GMT recommendation https://www.userightai.com/models/muse-spark chg_llama_5_open_weights_2026_04_08 Meta released Llama 5 on April 8, 2026: fully open weights, ~600B parameters, a 5M-token context window (the largest of any 2026 frontier release), and a Recursive Self-Improvement capability that lets it re-check its own reasoning mid-task. Hosted API deployments (Together AI, Fireworks, Groq, Bedrock) were announced but no major host had published verified per-token Llama 5 pricing as of August 6, 2026 — so, per our no-guessed-numbers policy, Llama 5 is tracked here and joins the ranked catalog when a real price lands. Meta's closed sibling Muse Spark, which does have verified pricing, has a full model page. Wed, 08 Apr 2026 00:00:00 GMT new-model https://www.userightai.com/models/gpt-5-4 chg_1 OpenAI's newer flagship model was added and immediately moved into the top overall spot for coding-heavy, decision-heavy workflows. Fri, 20 Mar 2026 00:00:00 GMT new-model https://www.userightai.com/models/gpt-5-4 chg_2 The coding leaderboard now favors GPT-5.4 over GPT-5.2 after refreshing the centralized recommendation scores. Fri, 20 Mar 2026 00:00:00 GMT recommendation https://www.userightai.com/models/gemini-2-5-flash chg_3 Budget recommendation strength improved after the model was updated with a larger context window in the shared dataset. Wed, 18 Mar 2026 00:00:00 GMT capability https://www.userightai.com/models/claude-4-sonnet chg_4 Claude 4 Sonnet kept the writing lead after reviewing tone consistency, revision quality, and long-form readability. Tue, 17 Mar 2026 00:00:00 GMT recommendation https://www.userightai.com/models/codestral-25-01 chg_5 The directory expanded affordable coding coverage by adding a stronger specialist option for engineering teams. Mon, 16 Mar 2026 00:00:00 GMT new-model https://www.userightai.com/models/gemini-2-5-pro chg_6 Research and long-context rankings were refreshed after re-evaluating synthesis quality across large inputs. Sun, 15 Mar 2026 00:00:00 GMT recommendation https://www.userightai.com/models/llama-4-scout chg_7 Scout moved into the stronger value tier after comparing affordable long-context options. Sat, 14 Mar 2026 00:00:00 GMT pricing https://www.userightai.com/models/claude-4-haiku chg_8 Haiku was reclassified as very fast after a fresh pass on quick-turn content workflows. Fri, 13 Mar 2026 00:00:00 GMT capability