[{"data":1,"prerenderedAt":1039},["ShallowReactive",2],{"post-claude-opus-4-7-launch-overview":3},{"id":4,"title":5,"author":6,"body":7,"category":1020,"cover":1021,"date":1022,"description":1023,"extension":1024,"featured":344,"meta":1025,"navigation":344,"path":1026,"published":344,"readingTime":1027,"seo":1028,"sitemap":1029,"stem":1030,"tags":1031,"updated":1022,"__hash__":1038},"posts/posts/claude-opus-4-7-launch-overview.md","Claude Opus 4.7 Is Here: What's New in Anthropic's April 2026 Release","Manash Sonowal",{"type":8,"value":9,"toc":1009},"minimark",[10,19,22,27,38,98,101,105,108,114,120,123,138,145,149,152,167,170,190,193,197,204,207,218,222,225,228,232,242,245,273,276,293,296,300,306,932,935,950,953,957,960,992,995,999,1002,1005],[11,12,13,14,18],"p",{},"Anthropic shipped ",[15,16,17],"strong",{},"Claude Opus 4.7"," today, April 16, 2026. It's a point release on paper — the pricing didn't move, the API shape didn't move, the context window didn't move — but the numbers under the hood did. Coding scores jumped, vision got a real capacity upgrade, and tool-use reliability took a noticeable step forward.",[11,20,21],{},"If you're already shipping on Opus 4.6, this is the kind of release where you change one string in your config and probably notice it in production the same afternoon. Let's walk through what actually changed.",[23,24,26],"h2",{"id":25},"the-headline-numbers","The headline numbers",[11,28,29,30,37],{},"Straight from Anthropic's ",[31,32,36],"a",{"href":33,"rel":34},"https://www.anthropic.com/claude/opus",[35],"nofollow","Opus product page"," and the launch materials:",[39,40,41,48,54,60,66,75,85,91],"ul",{},[42,43,44,47],"li",{},[15,45,46],{},"1M-token context window",", hybrid reasoning with adaptive thinking",[42,49,50,53],{},[15,51,52],{},"+13%"," lift over Opus 4.6 on a 93-task internal coding benchmark",[42,55,56,59],{},[15,57,58],{},"70% → from 58%"," on CursorBench (real repo-scale coding tasks)",[42,61,62,65],{},[15,63,64],{},"~3×"," more production tasks resolved end-to-end vs 4.6",[42,67,68,71,72],{},[15,69,70],{},"+14%"," on multi-step agentic workflows, with ",[15,73,74],{},"~2/3 fewer tool-use errors",[42,76,77,80,81,84],{},[15,78,79],{},"98.5%"," on visual-acuity benchmarks; image inputs up to ",[15,82,83],{},"2,576 px"," on the long edge (over 3× prior Claude models)",[42,86,87,90],{},[15,88,89],{},"0.715"," on a research-agent efficiency benchmark",[42,92,93,94,97],{},"Pricing ",[15,95,96],{},"unchanged",": $5 / M input, $25 / M output",[11,99,100],{},"The pattern here isn't \"a new paradigm\" — it's \"the same model family, measurably less likely to mess up.\" That's often the more useful kind of release.",[23,102,104],{"id":103},"smarter-coding","Smarter coding",[11,106,107],{},"The coding story is the loudest part of this launch. Two numbers to pay attention to:",[11,109,110,113],{},[15,111,112],{},"CursorBench went from 58% to 70%."," CursorBench is a repo-scale evaluation — not toy LeetCode, not isolated functions, but realistic multi-file edits where the model has to reason about a codebase it didn't write. A 12-point jump at that level is the difference between a model that sometimes needs hand-holding and one that can be trusted with a ticket.",[11,115,116,119],{},[15,117,118],{},"Anthropic reports ~3× more production tasks resolved end-to-end."," That framing matters. \"Resolved end-to-end\" means the output was good enough to ship, not just good enough to compile. If you've ever prompted a model, skimmed the diff, and quietly rewritten half of it — this is the metric targeting that experience.",[11,121,122],{},"What this means day-to-day:",[39,124,125,128,135],{},[42,126,127],{},"Larger refactors land in fewer turns.",[42,129,130,134],{},[131,132,133],"code",{},"plan → edit → run tests → fix"," loops are shorter because tool errors are rarer (more on that below).",[42,136,137],{},"You can hand it more of a ticket and less of a scaffold.",[11,139,140,141,144],{},"If you're using Claude Code, the Claude API with tool use, or any third-party agent harness, changing the model string to ",[131,142,143],{},"claude-opus-4-7"," is the whole upgrade path.",[23,146,148],{"id":147},"vision-upgrade","Vision upgrade",[11,150,151],{},"The multimodal story is the sleeper feature. Two things changed:",[153,154,155,161],"ol",{},[42,156,157,160],{},[15,158,159],{},"Image resolution ceiling is now 2,576 px on the long edge",", which is more than 3× what prior Claude models accepted before downscaling. You can send screenshots, PDFs, and diagrams without the model losing small text or fine-grained lines.",[42,162,163,166],{},[15,164,165],{},"98.5% on visual-acuity benchmarks",", with Anthropic specifically calling out gains on chemical structures and technical diagrams.",[11,168,169],{},"Concretely, this unlocks:",[39,171,172,178,184],{},[42,173,174,177],{},[15,175,176],{},"UI screenshots",": reading a full 1440p design mock without cropping or tiling.",[42,179,180,183],{},[15,181,182],{},"OCR on documents",": invoices and forms with small-font footnotes where 4.6 would quietly hallucinate.",[42,185,186,189],{},[15,187,188],{},"Diagram understanding",": architecture drawings, ER diagrams, math on a whiteboard — the \"explain this picture to me\" cases that used to feel 80% reliable now feel closer to 95%.",[11,191,192],{},"If you previously built a preprocessing pipeline to split images before sending them to Claude, you can likely rip most of it out.",[23,194,196],{"id":195},"agentic-reliability","Agentic reliability",[11,198,199,200,203],{},"The number that jumps out to anyone who builds agent loops: ",[15,201,202],{},"tool errors down by about two-thirds",".",[11,205,206],{},"If you've shipped an agent on top of 4.6, you know the failure mode — the model invents a function name, or fills a required argument with the wrong type, or calls the right tool with swapped parameters. Each of those is a retry or a hard failure, and retries compound fast in multi-step workflows.",[11,208,209,210,213,214,217],{},"Cutting that failure rate by 2/3, combined with a ",[15,211,212],{},"+14% lift on multi-step workflows"," and a ",[15,215,216],{},"0.715 research-agent score"," (Anthropic calls this their strongest efficiency baseline yet), means agent runs finish more often and with fewer wasted tokens. For anyone paying a bill per tool-use round-trip, that's money back.",[23,219,221],{"id":220},"hybrid-reasoning-and-adaptive-thinking","Hybrid reasoning and adaptive thinking",[11,223,224],{},"Opus 4.7 keeps the hybrid reasoning mode from 4.6 — one knob, two behaviors. On easy requests it answers fast. On hard requests it spends more compute \"thinking\" before it speaks, and that budget is adaptive rather than a flat toggle.",[11,226,227],{},"The practical implication: you don't need to hand-pick between \"fast\" and \"reasoning\" modes per request. Send it the problem and let it decide. If you want a hard cap on latency or cost, you can still set one — but the default behavior tends to land in a sensible place.",[23,229,231],{"id":230},"pricing-and-where-you-can-run-it","Pricing and where you can run it",[11,233,234,235,237,238,241],{},"Pricing is ",[15,236,96],{}," from Opus 4.6: ",[15,239,240],{},"$5 per million input tokens, $25 per million output tokens",". That's the detail most teams care about because it turns the upgrade decision into a no-brainer.",[11,243,244],{},"Cost levers to lean on:",[39,246,247,257,267],{},[42,248,249,252,253,256],{},[15,250,251],{},"Prompt caching",": up to ",[15,254,255],{},"90%"," savings on repeated context. Long system prompts, large file reads, retrieval-augmented prompts — all good candidates.",[42,258,259,262,263,266],{},[15,260,261],{},"Batch processing",": ",[15,264,265],{},"50%"," off for non-interactive jobs.",[42,268,269,272],{},[15,270,271],{},"US-only inference",": available at 1.1× for teams with data-residency requirements.",[11,274,275],{},"Availability on day one:",[39,277,278,281,284,287,290],{},[42,279,280],{},"Claude Pro, Max, Team, and Enterprise plans",[42,282,283],{},"Claude Platform API",[42,285,286],{},"Amazon Bedrock",[42,288,289],{},"Google Cloud Vertex AI",[42,291,292],{},"Microsoft Foundry",[11,294,295],{},"So whichever cloud you're already on, you can get to 4.7 without leaving it.",[23,297,299],{"id":298},"getting-started","Getting started",[11,301,302,303,305],{},"The model ID is ",[131,304,143],{},". Here's a minimal example with prompt caching enabled — the shape you'd want for anything with a non-trivial system prompt. Pick your language:",[307,308,311,689],"code-tabs",{"ariaLabel":309,"default":310},"Claude Opus 4.7 request example","php",[312,313,315],"template",{"v-slot:php":314},"",[316,317,320],"pre",{"className":318,"code":319,"language":310,"meta":314,"style":314},"language-php shiki shiki-themes github-light","use Illuminate\\Support\\Facades\\Http;\n\n$response = Http::withHeaders([\n    'x-api-key' => env('ANTHROPIC_API_KEY'),\n    'anthropic-version' => '2023-06-01',\n    'content-type' => 'application/json',\n])->post('https://api.anthropic.com/v1/messages', [\n    'model' => 'claude-opus-4-7',\n    'max_tokens' => 1024,\n    'system' => [\n        [\n            'type' => 'text',\n            'text' => 'You are a senior backend engineer reviewing code for a '\n                .'production PostgreSQL-backed Laravel service. Prioritize '\n                .'correctness, then security, then performance.',\n            // Cache this instruction block — it rarely changes between calls.\n            'cache_control' => ['type' => 'ephemeral'],\n        ],\n    ],\n    'messages' => [\n        [\n            'role' => 'user',\n            'content' => 'Review the diff in the attached PR and flag any N+1 queries.',\n        ],\n    ],\n])->json();\n\necho $response['content'][0]['text'], PHP_EOL;\necho 'cached input tokens: ', $response['usage']['cache_read_input_tokens'], PHP_EOL;\n",[131,321,322,339,346,368,390,404,417,437,450,463,474,480,493,504,513,523,530,552,558,564,574,579,592,605,610,615,628,633,664],{"__ignoreMap":314},[323,324,327,331,335],"span",{"class":325,"line":326},"line",1,[323,328,330],{"class":329},"sD7c4","use",[323,332,334],{"class":333},"sYu0t"," Illuminate\\Support\\Facades\\Http",[323,336,338],{"class":337},"sgsFI",";\n",[323,340,342],{"class":325,"line":341},2,[323,343,345],{"emptyLinePlaceholder":344},true,"\n",[323,347,349,352,355,358,361,365],{"class":325,"line":348},3,[323,350,351],{"class":337},"$response ",[323,353,354],{"class":329},"=",[323,356,357],{"class":333}," Http",[323,359,360],{"class":329},"::",[323,362,364],{"class":363},"s7eDp","withHeaders",[323,366,367],{"class":337},"([\n",[323,369,371,375,378,381,384,387],{"class":325,"line":370},4,[323,372,374],{"class":373},"sYBdl","    'x-api-key'",[323,376,377],{"class":329}," =>",[323,379,380],{"class":363}," env",[323,382,383],{"class":337},"(",[323,385,386],{"class":373},"'ANTHROPIC_API_KEY'",[323,388,389],{"class":337},"),\n",[323,391,393,396,398,401],{"class":325,"line":392},5,[323,394,395],{"class":373},"    'anthropic-version'",[323,397,377],{"class":329},[323,399,400],{"class":373}," '2023-06-01'",[323,402,403],{"class":337},",\n",[323,405,407,410,412,415],{"class":325,"line":406},6,[323,408,409],{"class":373},"    'content-type'",[323,411,377],{"class":329},[323,413,414],{"class":373}," 'application/json'",[323,416,403],{"class":337},[323,418,420,423,426,429,431,434],{"class":325,"line":419},7,[323,421,422],{"class":337},"])",[323,424,425],{"class":329},"->",[323,427,428],{"class":363},"post",[323,430,383],{"class":337},[323,432,433],{"class":373},"'https://api.anthropic.com/v1/messages'",[323,435,436],{"class":337},", [\n",[323,438,440,443,445,448],{"class":325,"line":439},8,[323,441,442],{"class":373},"    'model'",[323,444,377],{"class":329},[323,446,447],{"class":373}," 'claude-opus-4-7'",[323,449,403],{"class":337},[323,451,453,456,458,461],{"class":325,"line":452},9,[323,454,455],{"class":373},"    'max_tokens'",[323,457,377],{"class":329},[323,459,460],{"class":333}," 1024",[323,462,403],{"class":337},[323,464,466,469,471],{"class":325,"line":465},10,[323,467,468],{"class":373},"    'system'",[323,470,377],{"class":329},[323,472,473],{"class":337}," [\n",[323,475,477],{"class":325,"line":476},11,[323,478,479],{"class":337},"        [\n",[323,481,483,486,488,491],{"class":325,"line":482},12,[323,484,485],{"class":373},"            'type'",[323,487,377],{"class":329},[323,489,490],{"class":373}," 'text'",[323,492,403],{"class":337},[323,494,496,499,501],{"class":325,"line":495},13,[323,497,498],{"class":373},"            'text'",[323,500,377],{"class":329},[323,502,503],{"class":373}," 'You are a senior backend engineer reviewing code for a '\n",[323,505,507,510],{"class":325,"line":506},14,[323,508,509],{"class":329},"                .",[323,511,512],{"class":373},"'production PostgreSQL-backed Laravel service. Prioritize '\n",[323,514,516,518,521],{"class":325,"line":515},15,[323,517,509],{"class":329},[323,519,520],{"class":373},"'correctness, then security, then performance.'",[323,522,403],{"class":337},[323,524,526],{"class":325,"line":525},16,[323,527,529],{"class":528},"sAwPA","            // Cache this instruction block — it rarely changes between calls.\n",[323,531,533,536,538,541,544,546,549],{"class":325,"line":532},17,[323,534,535],{"class":373},"            'cache_control'",[323,537,377],{"class":329},[323,539,540],{"class":337}," [",[323,542,543],{"class":373},"'type'",[323,545,377],{"class":329},[323,547,548],{"class":373}," 'ephemeral'",[323,550,551],{"class":337},"],\n",[323,553,555],{"class":325,"line":554},18,[323,556,557],{"class":337},"        ],\n",[323,559,561],{"class":325,"line":560},19,[323,562,563],{"class":337},"    ],\n",[323,565,567,570,572],{"class":325,"line":566},20,[323,568,569],{"class":373},"    'messages'",[323,571,377],{"class":329},[323,573,473],{"class":337},[323,575,577],{"class":325,"line":576},21,[323,578,479],{"class":337},[323,580,582,585,587,590],{"class":325,"line":581},22,[323,583,584],{"class":373},"            'role'",[323,586,377],{"class":329},[323,588,589],{"class":373}," 'user'",[323,591,403],{"class":337},[323,593,595,598,600,603],{"class":325,"line":594},23,[323,596,597],{"class":373},"            'content'",[323,599,377],{"class":329},[323,601,602],{"class":373}," 'Review the diff in the attached PR and flag any N+1 queries.'",[323,604,403],{"class":337},[323,606,608],{"class":325,"line":607},24,[323,609,557],{"class":337},[323,611,613],{"class":325,"line":612},25,[323,614,563],{"class":337},[323,616,618,620,622,625],{"class":325,"line":617},26,[323,619,422],{"class":337},[323,621,425],{"class":329},[323,623,624],{"class":363},"json",[323,626,627],{"class":337},"();\n",[323,629,631],{"class":325,"line":630},27,[323,632,345],{"emptyLinePlaceholder":344},[323,634,636,639,642,645,648,651,653,656,659,662],{"class":325,"line":635},28,[323,637,638],{"class":333},"echo",[323,640,641],{"class":337}," $response[",[323,643,644],{"class":373},"'content'",[323,646,647],{"class":337},"][",[323,649,650],{"class":333},"0",[323,652,647],{"class":337},[323,654,655],{"class":373},"'text'",[323,657,658],{"class":337},"], ",[323,660,661],{"class":333},"PHP_EOL",[323,663,338],{"class":337},[323,665,667,669,672,675,678,680,683,685,687],{"class":325,"line":666},29,[323,668,638],{"class":333},[323,670,671],{"class":373}," 'cached input tokens: '",[323,673,674],{"class":337},", $response[",[323,676,677],{"class":373},"'usage'",[323,679,647],{"class":337},[323,681,682],{"class":373},"'cache_read_input_tokens'",[323,684,658],{"class":337},[323,686,661],{"class":333},[323,688,338],{"class":337},[312,690,691],{"v-slot:python":314},[316,692,696],{"className":693,"code":694,"language":695,"meta":314,"style":314},"language-python shiki shiki-themes github-light","from anthropic import Anthropic\n\nclient = Anthropic()\n\nresponse = client.messages.create(\n    model=\"claude-opus-4-7\",\n    max_tokens=1024,\n    system=[\n        {\n            \"type\": \"text\",\n            \"text\": (\n                \"You are a senior backend engineer reviewing code for a \"\n                \"production PostgreSQL-backed Django service. Prioritize \"\n                \"correctness, then security, then performance.\"\n            ),\n            # Cache this instruction block — it rarely changes between calls.\n            \"cache_control\": {\"type\": \"ephemeral\"},\n        }\n    ],\n    messages=[\n        {\n            \"role\": \"user\",\n            \"content\": \"Review the diff in the attached PR and flag any N+1 queries.\",\n        }\n    ],\n)\n\nprint(response.content[0].text)\nprint(\"cached input tokens:\", response.usage.cache_read_input_tokens)\n","python",[131,697,698,712,716,726,730,740,753,765,775,780,792,800,805,810,815,820,825,844,849,853,862,866,878,890,894,898,903,907,920],{"__ignoreMap":314},[323,699,700,703,706,709],{"class":325,"line":326},[323,701,702],{"class":329},"from",[323,704,705],{"class":337}," anthropic ",[323,707,708],{"class":329},"import",[323,710,711],{"class":337}," Anthropic\n",[323,713,714],{"class":325,"line":341},[323,715,345],{"emptyLinePlaceholder":344},[323,717,718,721,723],{"class":325,"line":348},[323,719,720],{"class":337},"client ",[323,722,354],{"class":329},[323,724,725],{"class":337}," Anthropic()\n",[323,727,728],{"class":325,"line":370},[323,729,345],{"emptyLinePlaceholder":344},[323,731,732,735,737],{"class":325,"line":392},[323,733,734],{"class":337},"response ",[323,736,354],{"class":329},[323,738,739],{"class":337}," client.messages.create(\n",[323,741,742,746,748,751],{"class":325,"line":406},[323,743,745],{"class":744},"sqxcx","    model",[323,747,354],{"class":329},[323,749,750],{"class":373},"\"claude-opus-4-7\"",[323,752,403],{"class":337},[323,754,755,758,760,763],{"class":325,"line":419},[323,756,757],{"class":744},"    max_tokens",[323,759,354],{"class":329},[323,761,762],{"class":333},"1024",[323,764,403],{"class":337},[323,766,767,770,772],{"class":325,"line":439},[323,768,769],{"class":744},"    system",[323,771,354],{"class":329},[323,773,774],{"class":337},"[\n",[323,776,777],{"class":325,"line":452},[323,778,779],{"class":337},"        {\n",[323,781,782,785,787,790],{"class":325,"line":465},[323,783,784],{"class":373},"            \"type\"",[323,786,262],{"class":337},[323,788,789],{"class":373},"\"text\"",[323,791,403],{"class":337},[323,793,794,797],{"class":325,"line":476},[323,795,796],{"class":373},"            \"text\"",[323,798,799],{"class":337},": (\n",[323,801,802],{"class":325,"line":482},[323,803,804],{"class":373},"                \"You are a senior backend engineer reviewing code for a \"\n",[323,806,807],{"class":325,"line":495},[323,808,809],{"class":373},"                \"production PostgreSQL-backed Django service. Prioritize \"\n",[323,811,812],{"class":325,"line":506},[323,813,814],{"class":373},"                \"correctness, then security, then performance.\"\n",[323,816,817],{"class":325,"line":515},[323,818,819],{"class":337},"            ),\n",[323,821,822],{"class":325,"line":525},[323,823,824],{"class":528},"            # Cache this instruction block — it rarely changes between calls.\n",[323,826,827,830,833,836,838,841],{"class":325,"line":532},[323,828,829],{"class":373},"            \"cache_control\"",[323,831,832],{"class":337},": {",[323,834,835],{"class":373},"\"type\"",[323,837,262],{"class":337},[323,839,840],{"class":373},"\"ephemeral\"",[323,842,843],{"class":337},"},\n",[323,845,846],{"class":325,"line":554},[323,847,848],{"class":337},"        }\n",[323,850,851],{"class":325,"line":560},[323,852,563],{"class":337},[323,854,855,858,860],{"class":325,"line":566},[323,856,857],{"class":744},"    messages",[323,859,354],{"class":329},[323,861,774],{"class":337},[323,863,864],{"class":325,"line":576},[323,865,779],{"class":337},[323,867,868,871,873,876],{"class":325,"line":581},[323,869,870],{"class":373},"            \"role\"",[323,872,262],{"class":337},[323,874,875],{"class":373},"\"user\"",[323,877,403],{"class":337},[323,879,880,883,885,888],{"class":325,"line":594},[323,881,882],{"class":373},"            \"content\"",[323,884,262],{"class":337},[323,886,887],{"class":373},"\"Review the diff in the attached PR and flag any N+1 queries.\"",[323,889,403],{"class":337},[323,891,892],{"class":325,"line":607},[323,893,848],{"class":337},[323,895,896],{"class":325,"line":612},[323,897,563],{"class":337},[323,899,900],{"class":325,"line":617},[323,901,902],{"class":337},")\n",[323,904,905],{"class":325,"line":630},[323,906,345],{"emptyLinePlaceholder":344},[323,908,909,912,915,917],{"class":325,"line":635},[323,910,911],{"class":333},"print",[323,913,914],{"class":337},"(response.content[",[323,916,650],{"class":333},[323,918,919],{"class":337},"].text)\n",[323,921,922,924,926,929],{"class":325,"line":666},[323,923,911],{"class":333},[323,925,383],{"class":337},[323,927,928],{"class":373},"\"cached input tokens:\"",[323,930,931],{"class":337},", response.usage.cache_read_input_tokens)\n",[11,933,934],{},"Two things to note:",[39,936,937,947],{},[42,938,939,942,943,946],{},[131,940,941],{},"cache_control"," on the system block is how you opt into prompt caching. On repeat calls with the same system text, ",[131,944,945],{},"cache_read_input_tokens"," climbs and your bill drops by up to 90% on that portion.",[42,948,949],{},"No other changes from your 4.6 setup. Same message schema, same tool-use format — whether you're hitting the API directly from PHP/Laravel or going through the official Python SDK.",[11,951,952],{},"If you're on Bedrock or Vertex, swap the transport and keep the model string.",[23,954,956],{"id":955},"should-you-upgrade-from-46","Should you upgrade from 4.6?",[11,958,959],{},"A short rubric:",[39,961,962,968,974,980,986],{},[42,963,964,967],{},[15,965,966],{},"You're doing code generation or repo-scale edits."," Yes. The CursorBench jump alone pays for the migration.",[42,969,970,973],{},[15,971,972],{},"You're running agents with tool use."," Yes. The 2/3 reduction in tool errors compounds across every multi-step run.",[42,975,976,979],{},[15,977,978],{},"You're doing vision-heavy work"," (screenshots, documents, diagrams). Yes. The resolution bump removes a whole class of preprocessing hacks.",[42,981,982,985],{},[15,983,984],{},"You're using Claude mostly for short Q&A or summarization."," Probably fine either way. You'll see small quality wins, but 4.6 isn't suddenly broken.",[42,987,988,991],{},[15,989,990],{},"You're cost-sensitive."," Same price as 4.6. Upgrade.",[11,993,994],{},"For most production workloads the answer is: flip the string, rerun your evals, ship.",[23,996,998],{"id":997},"closing","Closing",[11,1000,1001],{},"Opus 4.7 is the kind of release that rewards teams who already have a good eval harness — you'll see the gains clearly, and the upgrade cost is approximately zero. If you don't have evals yet, this is a good prompt to finally write them; you'll want to be able to measure the next point release the same way.",[11,1003,1004],{},"Thanks for reading!",[1006,1007,1008],"style",{},"html pre.shiki code .sD7c4, html code.shiki .sD7c4{--shiki-default:#D73A49}html pre.shiki code .sYu0t, html code.shiki .sYu0t{--shiki-default:#005CC5}html pre.shiki code .sgsFI, html code.shiki .sgsFI{--shiki-default:#24292E}html pre.shiki code .s7eDp, html code.shiki .s7eDp{--shiki-default:#6F42C1}html pre.shiki code .sYBdl, html code.shiki .sYBdl{--shiki-default:#032F62}html pre.shiki code .sAwPA, html code.shiki .sAwPA{--shiki-default:#6A737D}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html pre.shiki code .sqxcx, html code.shiki .sqxcx{--shiki-default:#E36209}",{"title":314,"searchDepth":341,"depth":341,"links":1010},[1011,1012,1013,1014,1015,1016,1017,1018,1019],{"id":25,"depth":341,"text":26},{"id":103,"depth":341,"text":104},{"id":147,"depth":341,"text":148},{"id":195,"depth":341,"text":196},{"id":220,"depth":341,"text":221},{"id":230,"depth":341,"text":231},{"id":298,"depth":341,"text":299},{"id":955,"depth":341,"text":956},{"id":997,"depth":341,"text":998},"AI & Machine Learning","/blog-covers/claude-opus-4-7-launch.png","2026-04-16","A quick developer tour of Claude Opus 4.7 — 1M context, sharper vision, +13% coding gains over 4.6, same $5/$25 pricing, and how to start using it today.","md",{},"/posts/claude-opus-4-7-launch-overview","9 min read",{"title":5,"description":1023},{"loc":1026,"lastmod":1022},"posts/claude-opus-4-7-launch-overview",[1032,1033,1034,1035,1036,1037],"claude","anthropic","llm","ai","opus-4-7","announcement","vY3EL88mrUx0u8lR0aOU7RXjm3p2cFm1ZKw1nSlpVFE",1777897734763]